1 /*
2 * Copyright (c) 1999, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "opto/addnode.hpp"
34 #include "opto/arraycopynode.hpp"
35 #include "opto/c2compiler.hpp"
36 #include "opto/callGenerator.hpp"
37 #include "opto/castnode.hpp"
38 #include "opto/cfgnode.hpp"
39 #include "opto/convertnode.hpp"
40 #include "opto/countbitsnode.hpp"
41 #include "opto/intrinsicnode.hpp"
42 #include "opto/idealKit.hpp"
43 #include "opto/mathexactnode.hpp"
44 #include "opto/movenode.hpp"
45 #include "opto/mulnode.hpp"
46 #include "opto/narrowptrnode.hpp"
47 #include "opto/opaquenode.hpp"
48 #include "opto/parse.hpp"
49 #include "opto/runtime.hpp"
50 #include "opto/subnode.hpp"
51 #include "prims/nativeLookup.hpp"
52 #include "prims/unsafe.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #ifdef TRACE_HAVE_INTRINSICS
55 #include "trace/traceMacros.hpp"
56 #endif
57
58 class LibraryIntrinsic : public InlineCallGenerator {
59 // Extend the set of intrinsics known to the runtime:
60 public:
61 private:
62 bool _is_virtual;
63 bool _does_virtual_dispatch;
64 int8_t _predicates_count; // Intrinsic is predicated by several conditions
65 int8_t _last_predicate; // Last generated predicate
66 vmIntrinsics::ID _intrinsic_id;
67
68 public:
69 LibraryIntrinsic(ciMethod* m, bool is_virtual, int predicates_count, bool does_virtual_dispatch, vmIntrinsics::ID id)
70 : InlineCallGenerator(m),
71 _is_virtual(is_virtual),
72 _does_virtual_dispatch(does_virtual_dispatch),
73 _predicates_count((int8_t)predicates_count),
74 _last_predicate((int8_t)-1),
75 _intrinsic_id(id)
76 {
77 }
78 virtual bool is_intrinsic() const { return true; }
79 virtual bool is_virtual() const { return _is_virtual; }
80 virtual bool is_predicated() const { return _predicates_count > 0; }
81 virtual int predicates_count() const { return _predicates_count; }
82 virtual bool does_virtual_dispatch() const { return _does_virtual_dispatch; }
83 virtual JVMState* generate(JVMState* jvms);
84 virtual Node* generate_predicate(JVMState* jvms, int predicate);
85 vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
86 };
87
88
89 // Local helper class for LibraryIntrinsic:
90 class LibraryCallKit : public GraphKit {
91 private:
92 LibraryIntrinsic* _intrinsic; // the library intrinsic being called
93 Node* _result; // the result node, if any
94 int _reexecute_sp; // the stack pointer when bytecode needs to be reexecuted
95
96 const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type);
97
98 public:
99 LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
100 : GraphKit(jvms),
101 _intrinsic(intrinsic),
102 _result(NULL)
103 {
104 // Check if this is a root compile. In that case we don't have a caller.
105 if (!jvms->has_method()) {
106 _reexecute_sp = sp();
107 } else {
108 // Find out how many arguments the interpreter needs when deoptimizing
109 // and save the stack pointer value so it can used by uncommon_trap.
110 // We find the argument count by looking at the declared signature.
111 bool ignored_will_link;
112 ciSignature* declared_signature = NULL;
113 ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
114 const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
115 _reexecute_sp = sp() + nargs; // "push" arguments back on stack
116 }
117 }
118
119 virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }
120
121 ciMethod* caller() const { return jvms()->method(); }
122 int bci() const { return jvms()->bci(); }
123 LibraryIntrinsic* intrinsic() const { return _intrinsic; }
124 vmIntrinsics::ID intrinsic_id() const { return _intrinsic->intrinsic_id(); }
125 ciMethod* callee() const { return _intrinsic->method(); }
126
127 bool try_to_inline(int predicate);
128 Node* try_to_predicate(int predicate);
129
130 void push_result() {
131 // Push the result onto the stack.
132 if (!stopped() && result() != NULL) {
133 BasicType bt = result()->bottom_type()->basic_type();
134 push_node(bt, result());
135 }
136 }
137
138 private:
139 void fatal_unexpected_iid(vmIntrinsics::ID iid) {
140 fatal("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid));
141 }
142
143 void set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
144 void set_result(RegionNode* region, PhiNode* value);
145 Node* result() { return _result; }
146
147 virtual int reexecute_sp() { return _reexecute_sp; }
148
149 // Helper functions to inline natives
150 Node* generate_guard(Node* test, RegionNode* region, float true_prob);
151 Node* generate_slow_guard(Node* test, RegionNode* region);
152 Node* generate_fair_guard(Node* test, RegionNode* region);
153 Node* generate_negative_guard(Node* index, RegionNode* region,
154 // resulting CastII of index:
155 Node* *pos_index = NULL);
156 Node* generate_limit_guard(Node* offset, Node* subseq_length,
157 Node* array_length,
158 RegionNode* region);
159 void generate_string_range_check(Node* array, Node* offset,
160 Node* length, bool char_count);
161 Node* generate_current_thread(Node* &tls_output);
162 Node* load_mirror_from_klass(Node* klass);
163 Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
164 RegionNode* region, int null_path,
165 int offset);
166 Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
167 RegionNode* region, int null_path) {
168 int offset = java_lang_Class::klass_offset_in_bytes();
169 return load_klass_from_mirror_common(mirror, never_see_null,
170 region, null_path,
171 offset);
172 }
173 Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
174 RegionNode* region, int null_path) {
175 int offset = java_lang_Class::array_klass_offset_in_bytes();
176 return load_klass_from_mirror_common(mirror, never_see_null,
177 region, null_path,
178 offset);
179 }
180 Node* generate_access_flags_guard(Node* kls,
181 int modifier_mask, int modifier_bits,
182 RegionNode* region);
183 Node* generate_interface_guard(Node* kls, RegionNode* region);
184 Node* generate_array_guard(Node* kls, RegionNode* region) {
185 return generate_array_guard_common(kls, region, false, false);
186 }
187 Node* generate_non_array_guard(Node* kls, RegionNode* region) {
188 return generate_array_guard_common(kls, region, false, true);
189 }
190 Node* generate_objArray_guard(Node* kls, RegionNode* region) {
191 return generate_array_guard_common(kls, region, true, false);
192 }
193 Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
194 return generate_array_guard_common(kls, region, true, true);
195 }
196 Node* generate_array_guard_common(Node* kls, RegionNode* region,
197 bool obj_array, bool not_array);
198 Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
199 CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
200 bool is_virtual = false, bool is_static = false);
201 CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
202 return generate_method_call(method_id, false, true);
203 }
204 CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
205 return generate_method_call(method_id, true, false);
206 }
207 Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
208 Node * field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static, ciInstanceKlass * fromKls);
209
210 Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae);
211 bool inline_string_compareTo(StrIntrinsicNode::ArgEnc ae);
212 bool inline_string_indexOf(StrIntrinsicNode::ArgEnc ae);
213 bool inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae);
214 Node* make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
215 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae);
216 bool inline_string_indexOfChar();
217 bool inline_string_equals(StrIntrinsicNode::ArgEnc ae);
218 bool inline_string_toBytesU();
219 bool inline_string_getCharsU();
220 bool inline_string_copy(bool compress);
221 bool inline_string_char_access(bool is_store);
222 Node* round_double_node(Node* n);
223 bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
224 bool inline_math_native(vmIntrinsics::ID id);
225 bool inline_math(vmIntrinsics::ID id);
226 template <typename OverflowOp>
227 bool inline_math_overflow(Node* arg1, Node* arg2);
228 void inline_math_mathExact(Node* math, Node* test);
229 bool inline_math_addExactI(bool is_increment);
230 bool inline_math_addExactL(bool is_increment);
231 bool inline_math_multiplyExactI();
232 bool inline_math_multiplyExactL();
233 bool inline_math_negateExactI();
234 bool inline_math_negateExactL();
235 bool inline_math_subtractExactI(bool is_decrement);
236 bool inline_math_subtractExactL(bool is_decrement);
237 bool inline_min_max(vmIntrinsics::ID id);
238 bool inline_notify(vmIntrinsics::ID id);
239 Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
240 // This returns Type::AnyPtr, RawPtr, or OopPtr.
241 int classify_unsafe_addr(Node* &base, Node* &offset);
242 Node* make_unsafe_address(Node* base, Node* offset);
243 // Helper for inline_unsafe_access.
244 // Generates the guards that check whether the result of
245 // Unsafe.getObject should be recorded in an SATB log buffer.
246 void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
247
248 typedef enum { Relaxed, Opaque, Volatile, Acquire, Release } AccessKind;
249 bool inline_unsafe_access(bool is_store, BasicType type, AccessKind kind, bool is_unaligned);
250 static bool klass_needs_init_guard(Node* kls);
251 bool inline_unsafe_allocate();
252 bool inline_unsafe_newArray(bool uninitialized);
253 bool inline_unsafe_copyMemory();
254 bool inline_native_currentThread();
255
256 bool inline_native_time_funcs(address method, const char* funcName);
257 #ifdef TRACE_HAVE_INTRINSICS
258 bool inline_native_classID();
259 bool inline_native_getBufferWriter();
260 #endif
261 bool inline_native_isInterrupted();
262 bool inline_native_Class_query(vmIntrinsics::ID id);
263 bool inline_native_subtype_check();
264 bool inline_native_getLength();
265 bool inline_array_copyOf(bool is_copyOfRange);
266 bool inline_array_equals(StrIntrinsicNode::ArgEnc ae);
267 bool inline_preconditions_checkIndex();
268 void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
269 bool inline_native_clone(bool is_virtual);
270 bool inline_native_Reflection_getCallerClass();
271 // Helper function for inlining native object hash method
272 bool inline_native_hashcode(bool is_virtual, bool is_static);
273 bool inline_native_getClass();
274
275 // Helper functions for inlining arraycopy
276 bool inline_arraycopy();
277 AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
278 RegionNode* slow_region);
279 JVMState* arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp);
280 void arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms, int saved_reexecute_sp,
281 uint new_idx);
282
283 typedef enum { LS_get_add, LS_get_set, LS_cmp_swap, LS_cmp_swap_weak, LS_cmp_exchange } LoadStoreKind;
284 MemNode::MemOrd access_kind_to_memord_LS(AccessKind access_kind, bool is_store);
285 MemNode::MemOrd access_kind_to_memord(AccessKind access_kind);
286 bool inline_unsafe_load_store(BasicType type, LoadStoreKind kind, AccessKind access_kind);
287 bool inline_unsafe_fence(vmIntrinsics::ID id);
288 bool inline_onspinwait();
289 bool inline_fp_conversions(vmIntrinsics::ID id);
290 bool inline_number_methods(vmIntrinsics::ID id);
291 bool inline_reference_get();
292 bool inline_Class_cast();
293 bool inline_aescrypt_Block(vmIntrinsics::ID id);
294 bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
295 bool inline_counterMode_AESCrypt(vmIntrinsics::ID id);
296 Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
297 Node* inline_counterMode_AESCrypt_predicate();
298 Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
299 Node* get_original_key_start_from_aescrypt_object(Node* aescrypt_object);
300 bool inline_ghash_processBlocks();
301 bool inline_sha_implCompress(vmIntrinsics::ID id);
302 bool inline_digestBase_implCompressMB(int predicate);
303 bool inline_sha_implCompressMB(Node* digestBaseObj, ciInstanceKlass* instklass_SHA,
304 bool long_state, address stubAddr, const char *stubName,
305 Node* src_start, Node* ofs, Node* limit);
306 Node* get_state_from_sha_object(Node *sha_object);
307 Node* get_state_from_sha5_object(Node *sha_object);
308 Node* inline_digestBase_implCompressMB_predicate(int predicate);
309 bool inline_encodeISOArray();
310 bool inline_updateCRC32();
311 bool inline_updateBytesCRC32();
312 bool inline_updateByteBufferCRC32();
313 Node* get_table_from_crc32c_class(ciInstanceKlass *crc32c_class);
314 bool inline_updateBytesCRC32C();
315 bool inline_updateDirectByteBufferCRC32C();
316 bool inline_updateBytesAdler32();
317 bool inline_updateByteBufferAdler32();
318 bool inline_multiplyToLen();
319 bool inline_hasNegatives();
320 bool inline_squareToLen();
321 bool inline_mulAdd();
322 bool inline_montgomeryMultiply();
323 bool inline_montgomerySquare();
324 bool inline_vectorizedMismatch();
325 bool inline_fma(vmIntrinsics::ID id);
326
327 bool inline_profileBoolean();
328 bool inline_isCompileConstant();
329 };
330
331 //---------------------------make_vm_intrinsic----------------------------
332 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
333 vmIntrinsics::ID id = m->intrinsic_id();
334 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
335
336 if (!m->is_loaded()) {
337 // Do not attempt to inline unloaded methods.
338 return NULL;
339 }
340
341 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
342 bool is_available = false;
343
344 {
345 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
346 // the compiler must transition to '_thread_in_vm' state because both
347 // methods access VM-internal data.
348 VM_ENTRY_MARK;
349 methodHandle mh(THREAD, m->get_Method());
350 is_available = compiler->is_intrinsic_supported(mh, is_virtual) &&
351 !C->directive()->is_intrinsic_disabled(mh) &&
352 !vmIntrinsics::is_disabled_by_flags(mh);
353
354 }
355
356 if (is_available) {
357 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
358 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
359 return new LibraryIntrinsic(m, is_virtual,
360 vmIntrinsics::predicates_needed(id),
361 vmIntrinsics::does_virtual_dispatch(id),
362 (vmIntrinsics::ID) id);
363 } else {
364 return NULL;
365 }
366 }
367
368 //----------------------register_library_intrinsics-----------------------
369 // Initialize this file's data structures, for each Compile instance.
370 void Compile::register_library_intrinsics() {
371 // Nothing to do here.
372 }
373
374 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
375 LibraryCallKit kit(jvms, this);
376 Compile* C = kit.C;
377 int nodes = C->unique();
378 #ifndef PRODUCT
379 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
380 char buf[1000];
381 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
382 tty->print_cr("Intrinsic %s", str);
383 }
384 #endif
385 ciMethod* callee = kit.callee();
386 const int bci = kit.bci();
387
388 // Try to inline the intrinsic.
389 if ((CheckIntrinsics ? callee->intrinsic_candidate() : true) &&
390 kit.try_to_inline(_last_predicate)) {
391 if (C->print_intrinsics() || C->print_inlining()) {
392 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
393 }
394 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
395 if (C->log()) {
396 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
397 vmIntrinsics::name_at(intrinsic_id()),
398 (is_virtual() ? " virtual='1'" : ""),
399 C->unique() - nodes);
400 }
401 // Push the result from the inlined method onto the stack.
402 kit.push_result();
403 C->print_inlining_update(this);
404 return kit.transfer_exceptions_into_jvms();
405 }
406
407 // The intrinsic bailed out
408 if (C->print_intrinsics() || C->print_inlining()) {
409 if (jvms->has_method()) {
410 // Not a root compile.
411 const char* msg;
412 if (callee->intrinsic_candidate()) {
413 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
414 } else {
415 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
416 : "failed to inline (intrinsic), method not annotated";
417 }
418 C->print_inlining(callee, jvms->depth() - 1, bci, msg);
419 } else {
420 // Root compile
421 tty->print("Did not generate intrinsic %s%s at bci:%d in",
422 vmIntrinsics::name_at(intrinsic_id()),
423 (is_virtual() ? " (virtual)" : ""), bci);
424 }
425 }
426 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
427 C->print_inlining_update(this);
428 return NULL;
429 }
430
431 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
432 LibraryCallKit kit(jvms, this);
433 Compile* C = kit.C;
434 int nodes = C->unique();
435 _last_predicate = predicate;
436 #ifndef PRODUCT
437 assert(is_predicated() && predicate < predicates_count(), "sanity");
438 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
439 char buf[1000];
440 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
441 tty->print_cr("Predicate for intrinsic %s", str);
442 }
443 #endif
444 ciMethod* callee = kit.callee();
445 const int bci = kit.bci();
446
447 Node* slow_ctl = kit.try_to_predicate(predicate);
448 if (!kit.failing()) {
449 if (C->print_intrinsics() || C->print_inlining()) {
450 C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual, predicate)" : "(intrinsic, predicate)");
451 }
452 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
453 if (C->log()) {
454 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
455 vmIntrinsics::name_at(intrinsic_id()),
456 (is_virtual() ? " virtual='1'" : ""),
457 C->unique() - nodes);
458 }
459 return slow_ctl; // Could be NULL if the check folds.
460 }
461
462 // The intrinsic bailed out
463 if (C->print_intrinsics() || C->print_inlining()) {
464 if (jvms->has_method()) {
465 // Not a root compile.
466 const char* msg = "failed to generate predicate for intrinsic";
467 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
468 } else {
469 // Root compile
470 C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
471 vmIntrinsics::name_at(intrinsic_id()),
472 (is_virtual() ? " (virtual)" : ""), bci);
473 }
474 }
475 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
476 return NULL;
477 }
478
479 bool LibraryCallKit::try_to_inline(int predicate) {
480 // Handle symbolic names for otherwise undistinguished boolean switches:
481 const bool is_store = true;
482 const bool is_compress = true;
483 const bool is_static = true;
484 const bool is_volatile = true;
485
486 if (!jvms()->has_method()) {
487 // Root JVMState has a null method.
488 assert(map()->memory()->Opcode() == Op_Parm, "");
489 // Insert the memory aliasing node
490 set_all_memory(reset_memory());
491 }
492 assert(merged_memory(), "");
493
494
495 switch (intrinsic_id()) {
496 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
497 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
498 case vmIntrinsics::_getClass: return inline_native_getClass();
499
500 case vmIntrinsics::_dsin:
501 case vmIntrinsics::_dcos:
502 case vmIntrinsics::_dtan:
503 case vmIntrinsics::_dabs:
504 case vmIntrinsics::_datan2:
505 case vmIntrinsics::_dsqrt:
506 case vmIntrinsics::_dexp:
507 case vmIntrinsics::_dlog:
508 case vmIntrinsics::_dlog10:
509 case vmIntrinsics::_dpow: return inline_math_native(intrinsic_id());
510
511 case vmIntrinsics::_min:
512 case vmIntrinsics::_max: return inline_min_max(intrinsic_id());
513
514 case vmIntrinsics::_notify:
515 case vmIntrinsics::_notifyAll:
516 if (InlineNotify) {
517 return inline_notify(intrinsic_id());
518 }
519 return false;
520
521 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
522 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
523 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
524 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
525 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
526 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
527 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
528 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
529 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
530 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
531 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
532 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
533
534 case vmIntrinsics::_arraycopy: return inline_arraycopy();
535
536 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
537 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
538 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
539 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
540
541 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
542 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
543 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
544 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
545 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
546 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
547 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar();
548
549 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
550 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
551
552 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
553 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
554 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
555 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
556
557 case vmIntrinsics::_compressStringC:
558 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
559 case vmIntrinsics::_inflateStringC:
560 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
561
562 case vmIntrinsics::_getObject: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
563 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
564 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
565 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
566 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
567 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
568 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
569 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
570 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
571
572 case vmIntrinsics::_putObject: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
573 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
574 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
575 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
576 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
577 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
578 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
579 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
580 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
581
582 case vmIntrinsics::_getObjectVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
583 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
584 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
585 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
586 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
587 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
588 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
589 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
590 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
591
592 case vmIntrinsics::_putObjectVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
593 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
594 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
595 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
596 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
597 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
598 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
599 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
600 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
601
602 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
603 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
604 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
605 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
606
607 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
608 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
609 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
610 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
611
612 case vmIntrinsics::_getObjectAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
613 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
614 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
615 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
616 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
617 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
618 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
619 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
620 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
621
622 case vmIntrinsics::_putObjectRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
623 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
624 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
625 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
626 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
627 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
628 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
629 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
630 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
631
632 case vmIntrinsics::_getObjectOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
633 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
634 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
635 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
636 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
637 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
638 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
639 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
640 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
641
642 case vmIntrinsics::_putObjectOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
643 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
644 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
645 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
646 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
647 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
648 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
649 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
650 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
651
652 case vmIntrinsics::_compareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
653 case vmIntrinsics::_compareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
654 case vmIntrinsics::_compareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
655 case vmIntrinsics::_compareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
656 case vmIntrinsics::_compareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
657
658 case vmIntrinsics::_weakCompareAndSwapObject: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
659 case vmIntrinsics::_weakCompareAndSwapObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
660 case vmIntrinsics::_weakCompareAndSwapObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
661 case vmIntrinsics::_weakCompareAndSwapObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
662 case vmIntrinsics::_weakCompareAndSwapByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
663 case vmIntrinsics::_weakCompareAndSwapByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
664 case vmIntrinsics::_weakCompareAndSwapByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
665 case vmIntrinsics::_weakCompareAndSwapByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
666 case vmIntrinsics::_weakCompareAndSwapShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
667 case vmIntrinsics::_weakCompareAndSwapShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
668 case vmIntrinsics::_weakCompareAndSwapShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
669 case vmIntrinsics::_weakCompareAndSwapShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
670 case vmIntrinsics::_weakCompareAndSwapInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
671 case vmIntrinsics::_weakCompareAndSwapIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
672 case vmIntrinsics::_weakCompareAndSwapIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
673 case vmIntrinsics::_weakCompareAndSwapIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
674 case vmIntrinsics::_weakCompareAndSwapLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
675 case vmIntrinsics::_weakCompareAndSwapLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
676 case vmIntrinsics::_weakCompareAndSwapLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
677 case vmIntrinsics::_weakCompareAndSwapLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
678
679 case vmIntrinsics::_compareAndExchangeObjectVolatile: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
680 case vmIntrinsics::_compareAndExchangeObjectAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
681 case vmIntrinsics::_compareAndExchangeObjectRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
682 case vmIntrinsics::_compareAndExchangeByteVolatile: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
683 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
684 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
685 case vmIntrinsics::_compareAndExchangeShortVolatile: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
686 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
687 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
688 case vmIntrinsics::_compareAndExchangeIntVolatile: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
689 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
690 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
691 case vmIntrinsics::_compareAndExchangeLongVolatile: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
692 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
693 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
694
695 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
696 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
697 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
698 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
699
700 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
701 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
702 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
703 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
704 case vmIntrinsics::_getAndSetObject: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
705
706 case vmIntrinsics::_loadFence:
707 case vmIntrinsics::_storeFence:
708 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
709
710 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
711
712 case vmIntrinsics::_currentThread: return inline_native_currentThread();
713 case vmIntrinsics::_isInterrupted: return inline_native_isInterrupted();
714
715 #ifdef TRACE_HAVE_INTRINSICS
716 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
717 case vmIntrinsics::_getClassId: return inline_native_classID();
718 case vmIntrinsics::_getBufferWriter: return inline_native_getBufferWriter();
719 #endif
720 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
721 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
722 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
723 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
724 case vmIntrinsics::_getLength: return inline_native_getLength();
725 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
726 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
727 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
728 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
729 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex();
730 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
731
732 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
733 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
734
735 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
736
737 case vmIntrinsics::_isInstance:
738 case vmIntrinsics::_getModifiers:
739 case vmIntrinsics::_isInterface:
740 case vmIntrinsics::_isArray:
741 case vmIntrinsics::_isPrimitive:
742 case vmIntrinsics::_getSuperclass:
743 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
744
745 case vmIntrinsics::_floatToRawIntBits:
746 case vmIntrinsics::_floatToIntBits:
747 case vmIntrinsics::_intBitsToFloat:
748 case vmIntrinsics::_doubleToRawLongBits:
749 case vmIntrinsics::_doubleToLongBits:
750 case vmIntrinsics::_longBitsToDouble: return inline_fp_conversions(intrinsic_id());
751
752 case vmIntrinsics::_numberOfLeadingZeros_i:
753 case vmIntrinsics::_numberOfLeadingZeros_l:
754 case vmIntrinsics::_numberOfTrailingZeros_i:
755 case vmIntrinsics::_numberOfTrailingZeros_l:
756 case vmIntrinsics::_bitCount_i:
757 case vmIntrinsics::_bitCount_l:
758 case vmIntrinsics::_reverseBytes_i:
759 case vmIntrinsics::_reverseBytes_l:
760 case vmIntrinsics::_reverseBytes_s:
761 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
762
763 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
764
765 case vmIntrinsics::_Reference_get: return inline_reference_get();
766
767 case vmIntrinsics::_Class_cast: return inline_Class_cast();
768
769 case vmIntrinsics::_aescrypt_encryptBlock:
770 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
771
772 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
773 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
774 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
775
776 case vmIntrinsics::_counterMode_AESCrypt:
777 return inline_counterMode_AESCrypt(intrinsic_id());
778
779 case vmIntrinsics::_sha_implCompress:
780 case vmIntrinsics::_sha2_implCompress:
781 case vmIntrinsics::_sha5_implCompress:
782 return inline_sha_implCompress(intrinsic_id());
783
784 case vmIntrinsics::_digestBase_implCompressMB:
785 return inline_digestBase_implCompressMB(predicate);
786
787 case vmIntrinsics::_multiplyToLen:
788 return inline_multiplyToLen();
789
790 case vmIntrinsics::_squareToLen:
791 return inline_squareToLen();
792
793 case vmIntrinsics::_mulAdd:
794 return inline_mulAdd();
795
796 case vmIntrinsics::_montgomeryMultiply:
797 return inline_montgomeryMultiply();
798 case vmIntrinsics::_montgomerySquare:
799 return inline_montgomerySquare();
800
801 case vmIntrinsics::_vectorizedMismatch:
802 return inline_vectorizedMismatch();
803
804 case vmIntrinsics::_ghash_processBlocks:
805 return inline_ghash_processBlocks();
806
807 case vmIntrinsics::_encodeISOArray:
808 case vmIntrinsics::_encodeByteISOArray:
809 return inline_encodeISOArray();
810
811 case vmIntrinsics::_updateCRC32:
812 return inline_updateCRC32();
813 case vmIntrinsics::_updateBytesCRC32:
814 return inline_updateBytesCRC32();
815 case vmIntrinsics::_updateByteBufferCRC32:
816 return inline_updateByteBufferCRC32();
817
818 case vmIntrinsics::_updateBytesCRC32C:
819 return inline_updateBytesCRC32C();
820 case vmIntrinsics::_updateDirectByteBufferCRC32C:
821 return inline_updateDirectByteBufferCRC32C();
822
823 case vmIntrinsics::_updateBytesAdler32:
824 return inline_updateBytesAdler32();
825 case vmIntrinsics::_updateByteBufferAdler32:
826 return inline_updateByteBufferAdler32();
827
828 case vmIntrinsics::_profileBoolean:
829 return inline_profileBoolean();
830 case vmIntrinsics::_isCompileConstant:
831 return inline_isCompileConstant();
832
833 case vmIntrinsics::_hasNegatives:
834 return inline_hasNegatives();
835
836 case vmIntrinsics::_fmaD:
837 case vmIntrinsics::_fmaF:
838 return inline_fma(intrinsic_id());
839
840 default:
841 // If you get here, it may be that someone has added a new intrinsic
842 // to the list in vmSymbols.hpp without implementing it here.
843 #ifndef PRODUCT
844 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
845 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
846 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
847 }
848 #endif
849 return false;
850 }
851 }
852
853 Node* LibraryCallKit::try_to_predicate(int predicate) {
854 if (!jvms()->has_method()) {
855 // Root JVMState has a null method.
856 assert(map()->memory()->Opcode() == Op_Parm, "");
857 // Insert the memory aliasing node
858 set_all_memory(reset_memory());
859 }
860 assert(merged_memory(), "");
861
862 switch (intrinsic_id()) {
863 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
864 return inline_cipherBlockChaining_AESCrypt_predicate(false);
865 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
866 return inline_cipherBlockChaining_AESCrypt_predicate(true);
867 case vmIntrinsics::_counterMode_AESCrypt:
868 return inline_counterMode_AESCrypt_predicate();
869 case vmIntrinsics::_digestBase_implCompressMB:
870 return inline_digestBase_implCompressMB_predicate(predicate);
871
872 default:
873 // If you get here, it may be that someone has added a new intrinsic
874 // to the list in vmSymbols.hpp without implementing it here.
875 #ifndef PRODUCT
876 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
877 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
878 vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
879 }
880 #endif
881 Node* slow_ctl = control();
882 set_control(top()); // No fast path instrinsic
883 return slow_ctl;
884 }
885 }
886
887 //------------------------------set_result-------------------------------
888 // Helper function for finishing intrinsics.
889 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
890 record_for_igvn(region);
891 set_control(_gvn.transform(region));
892 set_result( _gvn.transform(value));
893 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
894 }
895
896 //------------------------------generate_guard---------------------------
897 // Helper function for generating guarded fast-slow graph structures.
898 // The given 'test', if true, guards a slow path. If the test fails
899 // then a fast path can be taken. (We generally hope it fails.)
900 // In all cases, GraphKit::control() is updated to the fast path.
901 // The returned value represents the control for the slow path.
902 // The return value is never 'top'; it is either a valid control
903 // or NULL if it is obvious that the slow path can never be taken.
904 // Also, if region and the slow control are not NULL, the slow edge
905 // is appended to the region.
906 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
907 if (stopped()) {
908 // Already short circuited.
909 return NULL;
910 }
911
912 // Build an if node and its projections.
913 // If test is true we take the slow path, which we assume is uncommon.
914 if (_gvn.type(test) == TypeInt::ZERO) {
915 // The slow branch is never taken. No need to build this guard.
916 return NULL;
917 }
918
919 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
920
921 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
922 if (if_slow == top()) {
923 // The slow branch is never taken. No need to build this guard.
924 return NULL;
925 }
926
927 if (region != NULL)
928 region->add_req(if_slow);
929
930 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
931 set_control(if_fast);
932
933 return if_slow;
934 }
935
936 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
937 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
938 }
939 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
940 return generate_guard(test, region, PROB_FAIR);
941 }
942
943 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
944 Node* *pos_index) {
945 if (stopped())
946 return NULL; // already stopped
947 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
948 return NULL; // index is already adequately typed
949 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
950 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
951 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
952 if (is_neg != NULL && pos_index != NULL) {
953 // Emulate effect of Parse::adjust_map_after_if.
954 Node* ccast = new CastIINode(index, TypeInt::POS);
955 ccast->set_req(0, control());
956 (*pos_index) = _gvn.transform(ccast);
957 }
958 return is_neg;
959 }
960
961 // Make sure that 'position' is a valid limit index, in [0..length].
962 // There are two equivalent plans for checking this:
963 // A. (offset + copyLength) unsigned<= arrayLength
964 // B. offset <= (arrayLength - copyLength)
965 // We require that all of the values above, except for the sum and
966 // difference, are already known to be non-negative.
967 // Plan A is robust in the face of overflow, if offset and copyLength
968 // are both hugely positive.
969 //
970 // Plan B is less direct and intuitive, but it does not overflow at
971 // all, since the difference of two non-negatives is always
972 // representable. Whenever Java methods must perform the equivalent
973 // check they generally use Plan B instead of Plan A.
974 // For the moment we use Plan A.
975 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
976 Node* subseq_length,
977 Node* array_length,
978 RegionNode* region) {
979 if (stopped())
980 return NULL; // already stopped
981 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
982 if (zero_offset && subseq_length->eqv_uncast(array_length))
983 return NULL; // common case of whole-array copy
984 Node* last = subseq_length;
985 if (!zero_offset) // last += offset
986 last = _gvn.transform(new AddINode(last, offset));
987 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
988 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
989 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
990 return is_over;
991 }
992
993 // Emit range checks for the given String.value byte array
994 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
995 if (stopped()) {
996 return; // already stopped
997 }
998 RegionNode* bailout = new RegionNode(1);
999 record_for_igvn(bailout);
1000 if (char_count) {
1001 // Convert char count to byte count
1002 count = _gvn.transform(new LShiftINode(count, intcon(1)));
1003 }
1004
1005 // Offset and count must not be negative
1006 generate_negative_guard(offset, bailout);
1007 generate_negative_guard(count, bailout);
1008 // Offset + count must not exceed length of array
1009 generate_limit_guard(offset, count, load_array_length(array), bailout);
1010
1011 if (bailout->req() > 1) {
1012 PreserveJVMState pjvms(this);
1013 set_control(_gvn.transform(bailout));
1014 uncommon_trap(Deoptimization::Reason_intrinsic,
1015 Deoptimization::Action_maybe_recompile);
1016 }
1017 }
1018
1019 //--------------------------generate_current_thread--------------------
1020 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1021 ciKlass* thread_klass = env()->Thread_klass();
1022 const Type* thread_type = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1023 Node* thread = _gvn.transform(new ThreadLocalNode());
1024 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
1025 Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT, MemNode::unordered);
1026 tls_output = thread;
1027 return threadObj;
1028 }
1029
1030
1031 //------------------------------make_string_method_node------------------------
1032 // Helper method for String intrinsic functions. This version is called with
1033 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1034 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1035 // containing the lengths of str1 and str2.
1036 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1037 Node* result = NULL;
1038 switch (opcode) {
1039 case Op_StrIndexOf:
1040 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1041 str1_start, cnt1, str2_start, cnt2, ae);
1042 break;
1043 case Op_StrComp:
1044 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1045 str1_start, cnt1, str2_start, cnt2, ae);
1046 break;
1047 case Op_StrEquals:
1048 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1049 // Use the constant length if there is one because optimized match rule may exist.
1050 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1051 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1052 break;
1053 default:
1054 ShouldNotReachHere();
1055 return NULL;
1056 }
1057
1058 // All these intrinsics have checks.
1059 C->set_has_split_ifs(true); // Has chance for split-if optimization
1060
1061 return _gvn.transform(result);
1062 }
1063
1064 //------------------------------inline_string_compareTo------------------------
1065 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1066 Node* arg1 = argument(0);
1067 Node* arg2 = argument(1);
1068
1069 // Get start addr and length of first argument
1070 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1071 Node* arg1_cnt = load_array_length(arg1);
1072
1073 // Get start addr and length of second argument
1074 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1075 Node* arg2_cnt = load_array_length(arg2);
1076
1077 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1078 set_result(result);
1079 return true;
1080 }
1081
1082 //------------------------------inline_string_equals------------------------
1083 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1084 Node* arg1 = argument(0);
1085 Node* arg2 = argument(1);
1086
1087 // paths (plus control) merge
1088 RegionNode* region = new RegionNode(3);
1089 Node* phi = new PhiNode(region, TypeInt::BOOL);
1090
1091 if (!stopped()) {
1092 // Get start addr and length of first argument
1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1094 Node* arg1_cnt = load_array_length(arg1);
1095
1096 // Get start addr and length of second argument
1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1098 Node* arg2_cnt = load_array_length(arg2);
1099
1100 // Check for arg1_cnt != arg2_cnt
1101 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1102 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1103 Node* if_ne = generate_slow_guard(bol, NULL);
1104 if (if_ne != NULL) {
1105 phi->init_req(2, intcon(0));
1106 region->init_req(2, if_ne);
1107 }
1108
1109 // Check for count == 0 is done by assembler code for StrEquals.
1110
1111 if (!stopped()) {
1112 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1113 phi->init_req(1, equals);
1114 region->init_req(1, control());
1115 }
1116 }
1117
1118 // post merge
1119 set_control(_gvn.transform(region));
1120 record_for_igvn(region);
1121
1122 set_result(_gvn.transform(phi));
1123 return true;
1124 }
1125
1126 //------------------------------inline_array_equals----------------------------
1127 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1128 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1129 Node* arg1 = argument(0);
1130 Node* arg2 = argument(1);
1131
1132 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1133 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1134 return true;
1135 }
1136
1137 //------------------------------inline_hasNegatives------------------------------
1138 bool LibraryCallKit::inline_hasNegatives() {
1139 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1140 return false;
1141 }
1142
1143 assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1144 // no receiver since it is static method
1145 Node* ba = argument(0);
1146 Node* offset = argument(1);
1147 Node* len = argument(2);
1148
1149 // Range checks
1150 generate_string_range_check(ba, offset, len, false);
1151 if (stopped()) {
1152 return true;
1153 }
1154 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1155 Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1156 set_result(_gvn.transform(result));
1157 return true;
1158 }
1159
1160 bool LibraryCallKit::inline_preconditions_checkIndex() {
1161 Node* index = argument(0);
1162 Node* length = argument(1);
1163 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1164 return false;
1165 }
1166
1167 Node* len_pos_cmp = _gvn.transform(new CmpINode(length, intcon(0)));
1168 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1169
1170 {
1171 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1172 uncommon_trap(Deoptimization::Reason_intrinsic,
1173 Deoptimization::Action_make_not_entrant);
1174 }
1175
1176 if (stopped()) {
1177 return false;
1178 }
1179
1180 Node* rc_cmp = _gvn.transform(new CmpUNode(index, length));
1181 BoolTest::mask btest = BoolTest::lt;
1182 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1183 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1184 _gvn.set_type(rc, rc->Value(&_gvn));
1185 if (!rc_bool->is_Con()) {
1186 record_for_igvn(rc);
1187 }
1188 set_control(_gvn.transform(new IfTrueNode(rc)));
1189 {
1190 PreserveJVMState pjvms(this);
1191 set_control(_gvn.transform(new IfFalseNode(rc)));
1192 uncommon_trap(Deoptimization::Reason_range_check,
1193 Deoptimization::Action_make_not_entrant);
1194 }
1195
1196 if (stopped()) {
1197 return false;
1198 }
1199
1200 Node* result = new CastIINode(index, TypeInt::make(0, _gvn.type(length)->is_int()->_hi, Type::WidenMax));
1201 result->set_req(0, control());
1202 result = _gvn.transform(result);
1203 set_result(result);
1204 replace_in_map(index, result);
1205 return true;
1206 }
1207
1208 //------------------------------inline_string_indexOf------------------------
1209 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1210 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1211 return false;
1212 }
1213 Node* src = argument(0);
1214 Node* tgt = argument(1);
1215
1216 // Make the merge point
1217 RegionNode* result_rgn = new RegionNode(4);
1218 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1219
1220 // Get start addr and length of source string
1221 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1222 Node* src_count = load_array_length(src);
1223
1224 // Get start addr and length of substring
1225 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1226 Node* tgt_count = load_array_length(tgt);
1227
1228 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1229 // Divide src size by 2 if String is UTF16 encoded
1230 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1231 }
1232 if (ae == StrIntrinsicNode::UU) {
1233 // Divide substring size by 2 if String is UTF16 encoded
1234 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1235 }
1236
1237 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1238 if (result != NULL) {
1239 result_phi->init_req(3, result);
1240 result_rgn->init_req(3, control());
1241 }
1242 set_control(_gvn.transform(result_rgn));
1243 record_for_igvn(result_rgn);
1244 set_result(_gvn.transform(result_phi));
1245
1246 return true;
1247 }
1248
1249 //-----------------------------inline_string_indexOf-----------------------
1250 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1251 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1252 return false;
1253 }
1254 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1255 return false;
1256 }
1257 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1258 Node* src = argument(0); // byte[]
1259 Node* src_count = argument(1); // char count
1260 Node* tgt = argument(2); // byte[]
1261 Node* tgt_count = argument(3); // char count
1262 Node* from_index = argument(4); // char index
1263
1264 // Multiply byte array index by 2 if String is UTF16 encoded
1265 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1266 src_count = _gvn.transform(new SubINode(src_count, from_index));
1267 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1268 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1269
1270 // Range checks
1271 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1272 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1273 if (stopped()) {
1274 return true;
1275 }
1276
1277 RegionNode* region = new RegionNode(5);
1278 Node* phi = new PhiNode(region, TypeInt::INT);
1279
1280 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1281 if (result != NULL) {
1282 // The result is index relative to from_index if substring was found, -1 otherwise.
1283 // Generate code which will fold into cmove.
1284 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1285 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1286
1287 Node* if_lt = generate_slow_guard(bol, NULL);
1288 if (if_lt != NULL) {
1289 // result == -1
1290 phi->init_req(3, result);
1291 region->init_req(3, if_lt);
1292 }
1293 if (!stopped()) {
1294 result = _gvn.transform(new AddINode(result, from_index));
1295 phi->init_req(4, result);
1296 region->init_req(4, control());
1297 }
1298 }
1299
1300 set_control(_gvn.transform(region));
1301 record_for_igvn(region);
1302 set_result(_gvn.transform(phi));
1303
1304 return true;
1305 }
1306
1307 // Create StrIndexOfNode with fast path checks
1308 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1309 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1310 // Check for substr count > string count
1311 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1312 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1313 Node* if_gt = generate_slow_guard(bol, NULL);
1314 if (if_gt != NULL) {
1315 phi->init_req(1, intcon(-1));
1316 region->init_req(1, if_gt);
1317 }
1318 if (!stopped()) {
1319 // Check for substr count == 0
1320 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1321 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1322 Node* if_zero = generate_slow_guard(bol, NULL);
1323 if (if_zero != NULL) {
1324 phi->init_req(2, intcon(0));
1325 region->init_req(2, if_zero);
1326 }
1327 }
1328 if (!stopped()) {
1329 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1330 }
1331 return NULL;
1332 }
1333
1334 //-----------------------------inline_string_indexOfChar-----------------------
1335 bool LibraryCallKit::inline_string_indexOfChar() {
1336 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1337 return false;
1338 }
1339 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1340 return false;
1341 }
1342 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1343 Node* src = argument(0); // byte[]
1344 Node* tgt = argument(1); // tgt is int ch
1345 Node* from_index = argument(2);
1346 Node* max = argument(3);
1347
1348 Node* src_offset = _gvn.transform(new LShiftINode(from_index, intcon(1)));
1349 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1350 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1351
1352 // Range checks
1353 generate_string_range_check(src, src_offset, src_count, true);
1354 if (stopped()) {
1355 return true;
1356 }
1357
1358 RegionNode* region = new RegionNode(3);
1359 Node* phi = new PhiNode(region, TypeInt::INT);
1360
1361 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, StrIntrinsicNode::none);
1362 C->set_has_split_ifs(true); // Has chance for split-if optimization
1363 _gvn.transform(result);
1364
1365 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1366 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1367
1368 Node* if_lt = generate_slow_guard(bol, NULL);
1369 if (if_lt != NULL) {
1370 // result == -1
1371 phi->init_req(2, result);
1372 region->init_req(2, if_lt);
1373 }
1374 if (!stopped()) {
1375 result = _gvn.transform(new AddINode(result, from_index));
1376 phi->init_req(1, result);
1377 region->init_req(1, control());
1378 }
1379 set_control(_gvn.transform(region));
1380 record_for_igvn(region);
1381 set_result(_gvn.transform(phi));
1382
1383 return true;
1384 }
1385 //---------------------------inline_string_copy---------------------
1386 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1387 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1388 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1389 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1390 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1391 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1392 bool LibraryCallKit::inline_string_copy(bool compress) {
1393 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1394 return false;
1395 }
1396 int nargs = 5; // 2 oops, 3 ints
1397 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1398
1399 Node* src = argument(0);
1400 Node* src_offset = argument(1);
1401 Node* dst = argument(2);
1402 Node* dst_offset = argument(3);
1403 Node* length = argument(4);
1404
1405 // Check for allocation before we add nodes that would confuse
1406 // tightly_coupled_allocation()
1407 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1408
1409 // Figure out the size and type of the elements we will be copying.
1410 const Type* src_type = src->Value(&_gvn);
1411 const Type* dst_type = dst->Value(&_gvn);
1412 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1413 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1414 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1415 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1416 "Unsupported array types for inline_string_copy");
1417
1418 // Convert char[] offsets to byte[] offsets
1419 bool convert_src = (compress && src_elem == T_BYTE);
1420 bool convert_dst = (!compress && dst_elem == T_BYTE);
1421 if (convert_src) {
1422 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1423 } else if (convert_dst) {
1424 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1425 }
1426
1427 // Range checks
1428 generate_string_range_check(src, src_offset, length, convert_src);
1429 generate_string_range_check(dst, dst_offset, length, convert_dst);
1430 if (stopped()) {
1431 return true;
1432 }
1433
1434 Node* src_start = array_element_address(src, src_offset, src_elem);
1435 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1436 // 'src_start' points to src array + scaled offset
1437 // 'dst_start' points to dst array + scaled offset
1438 Node* count = NULL;
1439 if (compress) {
1440 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1441 } else {
1442 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1443 }
1444
1445 if (alloc != NULL) {
1446 if (alloc->maybe_set_complete(&_gvn)) {
1447 // "You break it, you buy it."
1448 InitializeNode* init = alloc->initialization();
1449 assert(init->is_complete(), "we just did this");
1450 init->set_complete_with_arraycopy();
1451 assert(dst->is_CheckCastPP(), "sanity");
1452 assert(dst->in(0)->in(0) == init, "dest pinned");
1453 }
1454 // Do not let stores that initialize this object be reordered with
1455 // a subsequent store that would make this object accessible by
1456 // other threads.
1457 // Record what AllocateNode this StoreStore protects so that
1458 // escape analysis can go from the MemBarStoreStoreNode to the
1459 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1460 // based on the escape status of the AllocateNode.
1461 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1462 }
1463 if (compress) {
1464 set_result(_gvn.transform(count));
1465 }
1466 return true;
1467 }
1468
1469 #ifdef _LP64
1470 #define XTOP ,top() /*additional argument*/
1471 #else //_LP64
1472 #define XTOP /*no additional argument*/
1473 #endif //_LP64
1474
1475 //------------------------inline_string_toBytesU--------------------------
1476 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1477 bool LibraryCallKit::inline_string_toBytesU() {
1478 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1479 return false;
1480 }
1481 // Get the arguments.
1482 Node* value = argument(0);
1483 Node* offset = argument(1);
1484 Node* length = argument(2);
1485
1486 Node* newcopy = NULL;
1487
1488 // Set the original stack and the reexecute bit for the interpreter to reexecute
1489 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1490 { PreserveReexecuteState preexecs(this);
1491 jvms()->set_should_reexecute(true);
1492
1493 // Check if a null path was taken unconditionally.
1494 value = null_check(value);
1495
1496 RegionNode* bailout = new RegionNode(1);
1497 record_for_igvn(bailout);
1498
1499 // Range checks
1500 generate_negative_guard(offset, bailout);
1501 generate_negative_guard(length, bailout);
1502 generate_limit_guard(offset, length, load_array_length(value), bailout);
1503 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1504 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1505
1506 if (bailout->req() > 1) {
1507 PreserveJVMState pjvms(this);
1508 set_control(_gvn.transform(bailout));
1509 uncommon_trap(Deoptimization::Reason_intrinsic,
1510 Deoptimization::Action_maybe_recompile);
1511 }
1512 if (stopped()) {
1513 return true;
1514 }
1515
1516 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1517 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1518 newcopy = new_array(klass_node, size, 0); // no arguments to push
1519 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy, NULL);
1520
1521 // Calculate starting addresses.
1522 Node* src_start = array_element_address(value, offset, T_CHAR);
1523 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1524
1525 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1526 const TypeInt* toffset = gvn().type(offset)->is_int();
1527 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1528
1529 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1530 const char* copyfunc_name = "arraycopy";
1531 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1532 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1533 OptoRuntime::fast_arraycopy_Type(),
1534 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1535 src_start, dst_start, ConvI2X(length) XTOP);
1536 // Do not let reads from the cloned object float above the arraycopy.
1537 if (alloc != NULL) {
1538 if (alloc->maybe_set_complete(&_gvn)) {
1539 // "You break it, you buy it."
1540 InitializeNode* init = alloc->initialization();
1541 assert(init->is_complete(), "we just did this");
1542 init->set_complete_with_arraycopy();
1543 assert(newcopy->is_CheckCastPP(), "sanity");
1544 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1545 }
1546 // Do not let stores that initialize this object be reordered with
1547 // a subsequent store that would make this object accessible by
1548 // other threads.
1549 // Record what AllocateNode this StoreStore protects so that
1550 // escape analysis can go from the MemBarStoreStoreNode to the
1551 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1552 // based on the escape status of the AllocateNode.
1553 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1554 } else {
1555 insert_mem_bar(Op_MemBarCPUOrder);
1556 }
1557 } // original reexecute is set back here
1558
1559 C->set_has_split_ifs(true); // Has chance for split-if optimization
1560 if (!stopped()) {
1561 set_result(newcopy);
1562 }
1563 return true;
1564 }
1565
1566 //------------------------inline_string_getCharsU--------------------------
1567 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1568 bool LibraryCallKit::inline_string_getCharsU() {
1569 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1570 return false;
1571 }
1572
1573 // Get the arguments.
1574 Node* src = argument(0);
1575 Node* src_begin = argument(1);
1576 Node* src_end = argument(2); // exclusive offset (i < src_end)
1577 Node* dst = argument(3);
1578 Node* dst_begin = argument(4);
1579
1580 // Check for allocation before we add nodes that would confuse
1581 // tightly_coupled_allocation()
1582 AllocateArrayNode* alloc = tightly_coupled_allocation(dst, NULL);
1583
1584 // Check if a null path was taken unconditionally.
1585 src = null_check(src);
1586 dst = null_check(dst);
1587 if (stopped()) {
1588 return true;
1589 }
1590
1591 // Get length and convert char[] offset to byte[] offset
1592 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1593 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1594
1595 // Range checks
1596 generate_string_range_check(src, src_begin, length, true);
1597 generate_string_range_check(dst, dst_begin, length, false);
1598 if (stopped()) {
1599 return true;
1600 }
1601
1602 if (!stopped()) {
1603 // Calculate starting addresses.
1604 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1605 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1606
1607 // Check if array addresses are aligned to HeapWordSize
1608 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1609 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1610 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1611 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1612
1613 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1614 const char* copyfunc_name = "arraycopy";
1615 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1616 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1617 OptoRuntime::fast_arraycopy_Type(),
1618 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1619 src_start, dst_start, ConvI2X(length) XTOP);
1620 // Do not let reads from the cloned object float above the arraycopy.
1621 if (alloc != NULL) {
1622 if (alloc->maybe_set_complete(&_gvn)) {
1623 // "You break it, you buy it."
1624 InitializeNode* init = alloc->initialization();
1625 assert(init->is_complete(), "we just did this");
1626 init->set_complete_with_arraycopy();
1627 assert(dst->is_CheckCastPP(), "sanity");
1628 assert(dst->in(0)->in(0) == init, "dest pinned");
1629 }
1630 // Do not let stores that initialize this object be reordered with
1631 // a subsequent store that would make this object accessible by
1632 // other threads.
1633 // Record what AllocateNode this StoreStore protects so that
1634 // escape analysis can go from the MemBarStoreStoreNode to the
1635 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1636 // based on the escape status of the AllocateNode.
1637 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
1638 } else {
1639 insert_mem_bar(Op_MemBarCPUOrder);
1640 }
1641 }
1642
1643 C->set_has_split_ifs(true); // Has chance for split-if optimization
1644 return true;
1645 }
1646
1647 //----------------------inline_string_char_access----------------------------
1648 // Store/Load char to/from byte[] array.
1649 // static void StringUTF16.putChar(byte[] val, int index, int c)
1650 // static char StringUTF16.getChar(byte[] val, int index)
1651 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1652 Node* value = argument(0);
1653 Node* index = argument(1);
1654 Node* ch = is_store ? argument(2) : NULL;
1655
1656 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1657 // correctly requires matched array shapes.
1658 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1659 "sanity: byte[] and char[] bases agree");
1660 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1661 "sanity: byte[] and char[] scales agree");
1662
1663 // Bail when getChar over constants is requested: constant folding would
1664 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1665 // Java method would constant fold nicely instead.
1666 if (!is_store && value->is_Con() && index->is_Con()) {
1667 return false;
1668 }
1669
1670 Node* adr = array_element_address(value, index, T_CHAR);
1671 if (is_store) {
1672 (void) store_to_memory(control(), adr, ch, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1673 false, false, true /* mismatched */);
1674 } else {
1675 ch = make_load(control(), adr, TypeInt::CHAR, T_CHAR, TypeAryPtr::BYTES, MemNode::unordered,
1676 LoadNode::DependsOnlyOnTest, false, false, true /* mismatched */);
1677 set_result(ch);
1678 }
1679 return true;
1680 }
1681
1682 //--------------------------round_double_node--------------------------------
1683 // Round a double node if necessary.
1684 Node* LibraryCallKit::round_double_node(Node* n) {
1685 if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
1686 n = _gvn.transform(new RoundDoubleNode(0, n));
1687 return n;
1688 }
1689
1690 //------------------------------inline_math-----------------------------------
1691 // public static double Math.abs(double)
1692 // public static double Math.sqrt(double)
1693 // public static double Math.log(double)
1694 // public static double Math.log10(double)
1695 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1696 Node* arg = round_double_node(argument(0));
1697 Node* n = NULL;
1698 switch (id) {
1699 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1700 case vmIntrinsics::_dsqrt: n = new SqrtDNode(C, control(), arg); break;
1701 default: fatal_unexpected_iid(id); break;
1702 }
1703 set_result(_gvn.transform(n));
1704 return true;
1705 }
1706
1707 //------------------------------runtime_math-----------------------------
1708 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1709 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1710 "must be (DD)D or (D)D type");
1711
1712 // Inputs
1713 Node* a = round_double_node(argument(0));
1714 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1715
1716 const TypePtr* no_memory_effects = NULL;
1717 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1718 no_memory_effects,
1719 a, top(), b, b ? top() : NULL);
1720 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1721 #ifdef ASSERT
1722 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1723 assert(value_top == top(), "second value must be top");
1724 #endif
1725
1726 set_result(value);
1727 return true;
1728 }
1729
1730 //------------------------------inline_math_native-----------------------------
1731 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1732 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1733 switch (id) {
1734 // These intrinsics are not properly supported on all hardware
1735 case vmIntrinsics::_dsin:
1736 return StubRoutines::dsin() != NULL ?
1737 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1738 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin), "SIN");
1739 case vmIntrinsics::_dcos:
1740 return StubRoutines::dcos() != NULL ?
1741 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1742 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos), "COS");
1743 case vmIntrinsics::_dtan:
1744 return StubRoutines::dtan() != NULL ?
1745 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1746 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1747 case vmIntrinsics::_dlog:
1748 return StubRoutines::dlog() != NULL ?
1749 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1750 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog), "LOG");
1751 case vmIntrinsics::_dlog10:
1752 return StubRoutines::dlog10() != NULL ?
1753 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1754 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1755
1756 // These intrinsics are supported on all hardware
1757 case vmIntrinsics::_dsqrt: return Matcher::match_rule_supported(Op_SqrtD) ? inline_math(id) : false;
1758 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_math(id) : false;
1759
1760 case vmIntrinsics::_dexp:
1761 return StubRoutines::dexp() != NULL ?
1762 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1763 runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp), "EXP");
1764 case vmIntrinsics::_dpow:
1765 return StubRoutines::dpow() != NULL ?
1766 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1767 runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow), "POW");
1768 #undef FN_PTR
1769
1770 // These intrinsics are not yet correctly implemented
1771 case vmIntrinsics::_datan2:
1772 return false;
1773
1774 default:
1775 fatal_unexpected_iid(id);
1776 return false;
1777 }
1778 }
1779
1780 static bool is_simple_name(Node* n) {
1781 return (n->req() == 1 // constant
1782 || (n->is_Type() && n->as_Type()->type()->singleton())
1783 || n->is_Proj() // parameter or return value
1784 || n->is_Phi() // local of some sort
1785 );
1786 }
1787
1788 //----------------------------inline_notify-----------------------------------*
1789 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1790 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1791 address func;
1792 if (id == vmIntrinsics::_notify) {
1793 func = OptoRuntime::monitor_notify_Java();
1794 } else {
1795 func = OptoRuntime::monitor_notifyAll_Java();
1796 }
1797 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1798 make_slow_call_ex(call, env()->Throwable_klass(), false);
1799 return true;
1800 }
1801
1802
1803 //----------------------------inline_min_max-----------------------------------
1804 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1805 set_result(generate_min_max(id, argument(0), argument(1)));
1806 return true;
1807 }
1808
1809 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1810 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1811 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1812 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1813 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1814
1815 {
1816 PreserveJVMState pjvms(this);
1817 PreserveReexecuteState preexecs(this);
1818 jvms()->set_should_reexecute(true);
1819
1820 set_control(slow_path);
1821 set_i_o(i_o());
1822
1823 uncommon_trap(Deoptimization::Reason_intrinsic,
1824 Deoptimization::Action_none);
1825 }
1826
1827 set_control(fast_path);
1828 set_result(math);
1829 }
1830
1831 template <typename OverflowOp>
1832 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1833 typedef typename OverflowOp::MathOp MathOp;
1834
1835 MathOp* mathOp = new MathOp(arg1, arg2);
1836 Node* operation = _gvn.transform( mathOp );
1837 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1838 inline_math_mathExact(operation, ofcheck);
1839 return true;
1840 }
1841
1842 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1843 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1844 }
1845
1846 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1847 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1848 }
1849
1850 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1851 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1852 }
1853
1854 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1855 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1856 }
1857
1858 bool LibraryCallKit::inline_math_negateExactI() {
1859 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1860 }
1861
1862 bool LibraryCallKit::inline_math_negateExactL() {
1863 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1864 }
1865
1866 bool LibraryCallKit::inline_math_multiplyExactI() {
1867 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1868 }
1869
1870 bool LibraryCallKit::inline_math_multiplyExactL() {
1871 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1872 }
1873
1874 Node*
1875 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1876 // These are the candidate return value:
1877 Node* xvalue = x0;
1878 Node* yvalue = y0;
1879
1880 if (xvalue == yvalue) {
1881 return xvalue;
1882 }
1883
1884 bool want_max = (id == vmIntrinsics::_max);
1885
1886 const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1887 const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1888 if (txvalue == NULL || tyvalue == NULL) return top();
1889 // This is not really necessary, but it is consistent with a
1890 // hypothetical MaxINode::Value method:
1891 int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1892
1893 // %%% This folding logic should (ideally) be in a different place.
1894 // Some should be inside IfNode, and there to be a more reliable
1895 // transformation of ?: style patterns into cmoves. We also want
1896 // more powerful optimizations around cmove and min/max.
1897
1898 // Try to find a dominating comparison of these guys.
1899 // It can simplify the index computation for Arrays.copyOf
1900 // and similar uses of System.arraycopy.
1901 // First, compute the normalized version of CmpI(x, y).
1902 int cmp_op = Op_CmpI;
1903 Node* xkey = xvalue;
1904 Node* ykey = yvalue;
1905 Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1906 if (ideal_cmpxy->is_Cmp()) {
1907 // E.g., if we have CmpI(length - offset, count),
1908 // it might idealize to CmpI(length, count + offset)
1909 cmp_op = ideal_cmpxy->Opcode();
1910 xkey = ideal_cmpxy->in(1);
1911 ykey = ideal_cmpxy->in(2);
1912 }
1913
1914 // Start by locating any relevant comparisons.
1915 Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1916 Node* cmpxy = NULL;
1917 Node* cmpyx = NULL;
1918 for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1919 Node* cmp = start_from->fast_out(k);
1920 if (cmp->outcnt() > 0 && // must have prior uses
1921 cmp->in(0) == NULL && // must be context-independent
1922 cmp->Opcode() == cmp_op) { // right kind of compare
1923 if (cmp->in(1) == xkey && cmp->in(2) == ykey) cmpxy = cmp;
1924 if (cmp->in(1) == ykey && cmp->in(2) == xkey) cmpyx = cmp;
1925 }
1926 }
1927
1928 const int NCMPS = 2;
1929 Node* cmps[NCMPS] = { cmpxy, cmpyx };
1930 int cmpn;
1931 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1932 if (cmps[cmpn] != NULL) break; // find a result
1933 }
1934 if (cmpn < NCMPS) {
1935 // Look for a dominating test that tells us the min and max.
1936 int depth = 0; // Limit search depth for speed
1937 Node* dom = control();
1938 for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1939 if (++depth >= 100) break;
1940 Node* ifproj = dom;
1941 if (!ifproj->is_Proj()) continue;
1942 Node* iff = ifproj->in(0);
1943 if (!iff->is_If()) continue;
1944 Node* bol = iff->in(1);
1945 if (!bol->is_Bool()) continue;
1946 Node* cmp = bol->in(1);
1947 if (cmp == NULL) continue;
1948 for (cmpn = 0; cmpn < NCMPS; cmpn++)
1949 if (cmps[cmpn] == cmp) break;
1950 if (cmpn == NCMPS) continue;
1951 BoolTest::mask btest = bol->as_Bool()->_test._test;
1952 if (ifproj->is_IfFalse()) btest = BoolTest(btest).negate();
1953 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1954 // At this point, we know that 'x btest y' is true.
1955 switch (btest) {
1956 case BoolTest::eq:
1957 // They are proven equal, so we can collapse the min/max.
1958 // Either value is the answer. Choose the simpler.
1959 if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1960 return yvalue;
1961 return xvalue;
1962 case BoolTest::lt: // x < y
1963 case BoolTest::le: // x <= y
1964 return (want_max ? yvalue : xvalue);
1965 case BoolTest::gt: // x > y
1966 case BoolTest::ge: // x >= y
1967 return (want_max ? xvalue : yvalue);
1968 }
1969 }
1970 }
1971
1972 // We failed to find a dominating test.
1973 // Let's pick a test that might GVN with prior tests.
1974 Node* best_bol = NULL;
1975 BoolTest::mask best_btest = BoolTest::illegal;
1976 for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1977 Node* cmp = cmps[cmpn];
1978 if (cmp == NULL) continue;
1979 for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
1980 Node* bol = cmp->fast_out(j);
1981 if (!bol->is_Bool()) continue;
1982 BoolTest::mask btest = bol->as_Bool()->_test._test;
1983 if (btest == BoolTest::eq || btest == BoolTest::ne) continue;
1984 if (cmp->in(1) == ykey) btest = BoolTest(btest).commute();
1985 if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
1986 best_bol = bol->as_Bool();
1987 best_btest = btest;
1988 }
1989 }
1990 }
1991
1992 Node* answer_if_true = NULL;
1993 Node* answer_if_false = NULL;
1994 switch (best_btest) {
1995 default:
1996 if (cmpxy == NULL)
1997 cmpxy = ideal_cmpxy;
1998 best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
1999 // and fall through:
2000 case BoolTest::lt: // x < y
2001 case BoolTest::le: // x <= y
2002 answer_if_true = (want_max ? yvalue : xvalue);
2003 answer_if_false = (want_max ? xvalue : yvalue);
2004 break;
2005 case BoolTest::gt: // x > y
2006 case BoolTest::ge: // x >= y
2007 answer_if_true = (want_max ? xvalue : yvalue);
2008 answer_if_false = (want_max ? yvalue : xvalue);
2009 break;
2010 }
2011
2012 jint hi, lo;
2013 if (want_max) {
2014 // We can sharpen the minimum.
2015 hi = MAX2(txvalue->_hi, tyvalue->_hi);
2016 lo = MAX2(txvalue->_lo, tyvalue->_lo);
2017 } else {
2018 // We can sharpen the maximum.
2019 hi = MIN2(txvalue->_hi, tyvalue->_hi);
2020 lo = MIN2(txvalue->_lo, tyvalue->_lo);
2021 }
2022
2023 // Use a flow-free graph structure, to avoid creating excess control edges
2024 // which could hinder other optimizations.
2025 // Since Math.min/max is often used with arraycopy, we want
2026 // tightly_coupled_allocation to be able to see beyond min/max expressions.
2027 Node* cmov = CMoveNode::make(NULL, best_bol,
2028 answer_if_false, answer_if_true,
2029 TypeInt::make(lo, hi, widen));
2030
2031 return _gvn.transform(cmov);
2032
2033 /*
2034 // This is not as desirable as it may seem, since Min and Max
2035 // nodes do not have a full set of optimizations.
2036 // And they would interfere, anyway, with 'if' optimizations
2037 // and with CMoveI canonical forms.
2038 switch (id) {
2039 case vmIntrinsics::_min:
2040 result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2041 case vmIntrinsics::_max:
2042 result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2043 default:
2044 ShouldNotReachHere();
2045 }
2046 */
2047 }
2048
2049 inline int
2050 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
2051 const TypePtr* base_type = TypePtr::NULL_PTR;
2052 if (base != NULL) base_type = _gvn.type(base)->isa_ptr();
2053 if (base_type == NULL) {
2054 // Unknown type.
2055 return Type::AnyPtr;
2056 } else if (base_type == TypePtr::NULL_PTR) {
2057 // Since this is a NULL+long form, we have to switch to a rawptr.
2058 base = _gvn.transform(new CastX2PNode(offset));
2059 offset = MakeConX(0);
2060 return Type::RawPtr;
2061 } else if (base_type->base() == Type::RawPtr) {
2062 return Type::RawPtr;
2063 } else if (base_type->isa_oopptr()) {
2064 // Base is never null => always a heap address.
2065 if (base_type->ptr() == TypePtr::NotNull) {
2066 return Type::OopPtr;
2067 }
2068 // Offset is small => always a heap address.
2069 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2070 if (offset_type != NULL &&
2071 base_type->offset() == 0 && // (should always be?)
2072 offset_type->_lo >= 0 &&
2073 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2074 return Type::OopPtr;
2075 }
2076 // Otherwise, it might either be oop+off or NULL+addr.
2077 return Type::AnyPtr;
2078 } else {
2079 // No information:
2080 return Type::AnyPtr;
2081 }
2082 }
2083
2084 inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
2085 int kind = classify_unsafe_addr(base, offset);
2086 if (kind == Type::RawPtr) {
2087 return basic_plus_adr(top(), base, offset);
2088 } else {
2089 return basic_plus_adr(base, offset);
2090 }
2091 }
2092
2093 //--------------------------inline_number_methods-----------------------------
2094 // inline int Integer.numberOfLeadingZeros(int)
2095 // inline int Long.numberOfLeadingZeros(long)
2096 //
2097 // inline int Integer.numberOfTrailingZeros(int)
2098 // inline int Long.numberOfTrailingZeros(long)
2099 //
2100 // inline int Integer.bitCount(int)
2101 // inline int Long.bitCount(long)
2102 //
2103 // inline char Character.reverseBytes(char)
2104 // inline short Short.reverseBytes(short)
2105 // inline int Integer.reverseBytes(int)
2106 // inline long Long.reverseBytes(long)
2107 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2108 Node* arg = argument(0);
2109 Node* n = NULL;
2110 switch (id) {
2111 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2112 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2113 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2114 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2115 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2116 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2117 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2118 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2119 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2120 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2121 default: fatal_unexpected_iid(id); break;
2122 }
2123 set_result(_gvn.transform(n));
2124 return true;
2125 }
2126
2127 //----------------------------inline_unsafe_access----------------------------
2128
2129 // Helper that guards and inserts a pre-barrier.
2130 void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2131 Node* pre_val, bool need_mem_bar) {
2132 // We could be accessing the referent field of a reference object. If so, when G1
2133 // is enabled, we need to log the value in the referent field in an SATB buffer.
2134 // This routine performs some compile time filters and generates suitable
2135 // runtime filters that guard the pre-barrier code.
2136 // Also add memory barrier for non volatile load from the referent field
2137 // to prevent commoning of loads across safepoint.
2138 if (!UseG1GC && !need_mem_bar)
2139 return;
2140
2141 // Some compile time checks.
2142
2143 // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
2144 const TypeX* otype = offset->find_intptr_t_type();
2145 if (otype != NULL && otype->is_con() &&
2146 otype->get_con() != java_lang_ref_Reference::referent_offset) {
2147 // Constant offset but not the reference_offset so just return
2148 return;
2149 }
2150
2151 // We only need to generate the runtime guards for instances.
2152 const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
2153 if (btype != NULL) {
2154 if (btype->isa_aryptr()) {
2155 // Array type so nothing to do
2156 return;
2157 }
2158
2159 const TypeInstPtr* itype = btype->isa_instptr();
2160 if (itype != NULL) {
2161 // Can the klass of base_oop be statically determined to be
2162 // _not_ a sub-class of Reference and _not_ Object?
2163 ciKlass* klass = itype->klass();
2164 if ( klass->is_loaded() &&
2165 !klass->is_subtype_of(env()->Reference_klass()) &&
2166 !env()->Object_klass()->is_subtype_of(klass)) {
2167 return;
2168 }
2169 }
2170 }
2171
2172 // The compile time filters did not reject base_oop/offset so
2173 // we need to generate the following runtime filters
2174 //
2175 // if (offset == java_lang_ref_Reference::_reference_offset) {
2176 // if (instance_of(base, java.lang.ref.Reference)) {
2177 // pre_barrier(_, pre_val, ...);
2178 // }
2179 // }
2180
2181 float likely = PROB_LIKELY( 0.999);
2182 float unlikely = PROB_UNLIKELY(0.999);
2183
2184 IdealKit ideal(this);
2185 #define __ ideal.
2186
2187 Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2188
2189 __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
2190 // Update graphKit memory and control from IdealKit.
2191 sync_kit(ideal);
2192
2193 Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
2194 Node* is_instof = gen_instanceof(base_oop, ref_klass_con);
2195
2196 // Update IdealKit memory and control from graphKit.
2197 __ sync_kit(this);
2198
2199 Node* one = __ ConI(1);
2200 // is_instof == 0 if base_oop == NULL
2201 __ if_then(is_instof, BoolTest::eq, one, unlikely); {
2202
2203 // Update graphKit from IdeakKit.
2204 sync_kit(ideal);
2205
2206 // Use the pre-barrier to record the value in the referent field
2207 pre_barrier(false /* do_load */,
2208 __ ctrl(),
2209 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2210 pre_val /* pre_val */,
2211 T_OBJECT);
2212 if (need_mem_bar) {
2213 // Add memory barrier to prevent commoning reads from this field
2214 // across safepoint since GC can change its value.
2215 insert_mem_bar(Op_MemBarCPUOrder);
2216 }
2217 // Update IdealKit from graphKit.
2218 __ sync_kit(this);
2219
2220 } __ end_if(); // _ref_type != ref_none
2221 } __ end_if(); // offset == referent_offset
2222
2223 // Final sync IdealKit and GraphKit.
2224 final_sync(ideal);
2225 #undef __
2226 }
2227
2228
2229 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2230 // Attempt to infer a sharper value type from the offset and base type.
2231 ciKlass* sharpened_klass = NULL;
2232
2233 // See if it is an instance field, with an object type.
2234 if (alias_type->field() != NULL) {
2235 if (alias_type->field()->type()->is_klass()) {
2236 sharpened_klass = alias_type->field()->type()->as_klass();
2237 }
2238 }
2239
2240 // See if it is a narrow oop array.
2241 if (adr_type->isa_aryptr()) {
2242 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2243 const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
2244 if (elem_type != NULL) {
2245 sharpened_klass = elem_type->klass();
2246 }
2247 }
2248 }
2249
2250 // The sharpened class might be unloaded if there is no class loader
2251 // contraint in place.
2252 if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2253 const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2254
2255 #ifndef PRODUCT
2256 if (C->print_intrinsics() || C->print_inlining()) {
2257 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2258 tty->print(" sharpened value: "); tjp->dump(); tty->cr();
2259 }
2260 #endif
2261 // Sharpen the value type.
2262 return tjp;
2263 }
2264 return NULL;
2265 }
2266
2267 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2268 if (callee()->is_static()) return false; // caller must have the capability!
2269 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2270 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2271 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2272
2273 #ifndef PRODUCT
2274 {
2275 ResourceMark rm;
2276 // Check the signatures.
2277 ciSignature* sig = callee()->signature();
2278 #ifdef ASSERT
2279 if (!is_store) {
2280 // Object getObject(Object base, int/long offset), etc.
2281 BasicType rtype = sig->return_type()->basic_type();
2282 assert(rtype == type, "getter must return the expected value");
2283 assert(sig->count() == 2, "oop getter has 2 arguments");
2284 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2285 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2286 } else {
2287 // void putObject(Object base, int/long offset, Object x), etc.
2288 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2289 assert(sig->count() == 3, "oop putter has 3 arguments");
2290 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2291 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2292 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2293 assert(vtype == type, "putter must accept the expected value");
2294 }
2295 #endif // ASSERT
2296 }
2297 #endif //PRODUCT
2298
2299 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2300
2301 Node* receiver = argument(0); // type: oop
2302
2303 // Build address expression.
2304 Node* adr;
2305 Node* heap_base_oop = top();
2306 Node* offset = top();
2307 Node* val;
2308
2309 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2310 Node* base = argument(1); // type: oop
2311 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2312 offset = argument(2); // type: long
2313 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2314 // to be plain byte offsets, which are also the same as those accepted
2315 // by oopDesc::field_base.
2316 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2317 "fieldOffset must be byte-scaled");
2318 // 32-bit machines ignore the high half!
2319 offset = ConvL2X(offset);
2320 adr = make_unsafe_address(base, offset);
2321 if (_gvn.type(base)->isa_ptr() != TypePtr::NULL_PTR) {
2322 heap_base_oop = base;
2323 } else if (type == T_OBJECT) {
2324 return false; // off-heap oop accesses are not supported
2325 }
2326
2327 // Can base be NULL? Otherwise, always on-heap access.
2328 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop));
2329
2330 val = is_store ? argument(4) : NULL;
2331
2332 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2333
2334 // Try to categorize the address.
2335 Compile::AliasType* alias_type = C->alias_type(adr_type);
2336 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2337
2338 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2339 alias_type->adr_type() == TypeAryPtr::RANGE) {
2340 return false; // not supported
2341 }
2342
2343 bool mismatched = false;
2344 BasicType bt = alias_type->basic_type();
2345 if (bt != T_ILLEGAL) {
2346 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2347 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2348 // Alias type doesn't differentiate between byte[] and boolean[]).
2349 // Use address type to get the element type.
2350 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2351 }
2352 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2353 // accessing an array field with getObject is not a mismatch
2354 bt = T_OBJECT;
2355 }
2356 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2357 // Don't intrinsify mismatched object accesses
2358 return false;
2359 }
2360 mismatched = (bt != type);
2361 } else if (alias_type->adr_type()->isa_oopptr()) {
2362 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2363 }
2364
2365 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2366
2367 // First guess at the value type.
2368 const Type *value_type = Type::get_const_basic_type(type);
2369
2370 // We will need memory barriers unless we can determine a unique
2371 // alias category for this reference. (Note: If for some reason
2372 // the barriers get omitted and the unsafe reference begins to "pollute"
2373 // the alias analysis of the rest of the graph, either Compile::can_alias
2374 // or Compile::must_alias will throw a diagnostic assert.)
2375 bool need_mem_bar;
2376 switch (kind) {
2377 case Relaxed:
2378 need_mem_bar = mismatched || can_access_non_heap;
2379 break;
2380 case Opaque:
2381 // Opaque uses CPUOrder membars for protection against code movement.
2382 case Acquire:
2383 case Release:
2384 case Volatile:
2385 need_mem_bar = true;
2386 break;
2387 default:
2388 ShouldNotReachHere();
2389 }
2390
2391 // Some accesses require access atomicity for all types, notably longs and doubles.
2392 // When AlwaysAtomicAccesses is enabled, all accesses are atomic.
2393 bool requires_atomic_access = false;
2394 switch (kind) {
2395 case Relaxed:
2396 requires_atomic_access = AlwaysAtomicAccesses;
2397 break;
2398 case Opaque:
2399 // Opaque accesses are atomic.
2400 case Acquire:
2401 case Release:
2402 case Volatile:
2403 requires_atomic_access = true;
2404 break;
2405 default:
2406 ShouldNotReachHere();
2407 }
2408
2409 // Figure out the memory ordering.
2410 // Acquire/Release/Volatile accesses require marking the loads/stores with MemOrd
2411 MemNode::MemOrd mo = access_kind_to_memord_LS(kind, is_store);
2412
2413 // If we are reading the value of the referent field of a Reference
2414 // object (either by using Unsafe directly or through reflection)
2415 // then, if G1 is enabled, we need to record the referent in an
2416 // SATB log buffer using the pre-barrier mechanism.
2417 // Also we need to add memory barrier to prevent commoning reads
2418 // from this field across safepoint since GC can change its value.
2419 bool need_read_barrier = !is_store &&
2420 offset != top() && heap_base_oop != top();
2421
2422 if (!is_store && type == T_OBJECT) {
2423 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2424 if (tjp != NULL) {
2425 value_type = tjp;
2426 }
2427 }
2428
2429 receiver = null_check(receiver);
2430 if (stopped()) {
2431 return true;
2432 }
2433 // Heap pointers get a null-check from the interpreter,
2434 // as a courtesy. However, this is not guaranteed by Unsafe,
2435 // and it is not possible to fully distinguish unintended nulls
2436 // from intended ones in this API.
2437
2438 // We need to emit leading and trailing CPU membars (see below) in
2439 // addition to memory membars for special access modes. This is a little
2440 // too strong, but avoids the need to insert per-alias-type
2441 // volatile membars (for stores; compare Parse::do_put_xxx), which
2442 // we cannot do effectively here because we probably only have a
2443 // rough approximation of type.
2444
2445 switch(kind) {
2446 case Relaxed:
2447 case Opaque:
2448 case Acquire:
2449 break;
2450 case Release:
2451 case Volatile:
2452 if (is_store) {
2453 insert_mem_bar(Op_MemBarRelease);
2454 } else {
2455 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2456 insert_mem_bar(Op_MemBarVolatile);
2457 }
2458 }
2459 break;
2460 default:
2461 ShouldNotReachHere();
2462 }
2463
2464 // Memory barrier to prevent normal and 'unsafe' accesses from
2465 // bypassing each other. Happens after null checks, so the
2466 // exception paths do not take memory state from the memory barrier,
2467 // so there's no problems making a strong assert about mixing users
2468 // of safe & unsafe memory.
2469 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2470
2471 if (!is_store) {
2472 Node* p = NULL;
2473 // Try to constant fold a load from a constant field
2474 ciField* field = alias_type->field();
2475 if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2476 // final or stable field
2477 p = make_constant_from_field(field, heap_base_oop);
2478 }
2479 if (p == NULL) {
2480 // To be valid, unsafe loads may depend on other conditions than
2481 // the one that guards them: pin the Load node
2482 p = make_load(control(), adr, value_type, type, adr_type, mo, LoadNode::Pinned, requires_atomic_access, unaligned, mismatched);
2483 // load value
2484 switch (type) {
2485 case T_BOOLEAN:
2486 {
2487 // Normalize the value returned by getBoolean in the following cases
2488 if (mismatched ||
2489 heap_base_oop == top() || // - heap_base_oop is NULL or
2490 (can_access_non_heap && alias_type->field() == NULL) // - heap_base_oop is potentially NULL
2491 // and the unsafe access is made to large offset
2492 // (i.e., larger than the maximum offset necessary for any
2493 // field access)
2494 ) {
2495 IdealKit ideal = IdealKit(this);
2496 #define __ ideal.
2497 IdealVariable normalized_result(ideal);
2498 __ declarations_done();
2499 __ set(normalized_result, p);
2500 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2501 __ set(normalized_result, ideal.ConI(1));
2502 ideal.end_if();
2503 final_sync(ideal);
2504 p = __ value(normalized_result);
2505 #undef __
2506 }
2507 }
2508 case T_CHAR:
2509 case T_BYTE:
2510 case T_SHORT:
2511 case T_INT:
2512 case T_LONG:
2513 case T_FLOAT:
2514 case T_DOUBLE:
2515 break;
2516 case T_OBJECT:
2517 if (need_read_barrier) {
2518 // We do not require a mem bar inside pre_barrier if need_mem_bar
2519 // is set: the barriers would be emitted by us.
2520 insert_pre_barrier(heap_base_oop, offset, p, !need_mem_bar);
2521 }
2522 break;
2523 case T_ADDRESS:
2524 // Cast to an int type.
2525 p = _gvn.transform(new CastP2XNode(NULL, p));
2526 p = ConvX2UL(p);
2527 break;
2528 default:
2529 fatal("unexpected type %d: %s", type, type2name(type));
2530 break;
2531 }
2532 }
2533 // The load node has the control of the preceding MemBarCPUOrder. All
2534 // following nodes will have the control of the MemBarCPUOrder inserted at
2535 // the end of this method. So, pushing the load onto the stack at a later
2536 // point is fine.
2537 set_result(p);
2538 } else {
2539 // place effect of store into memory
2540 switch (type) {
2541 case T_DOUBLE:
2542 val = dstore_rounding(val);
2543 break;
2544 case T_ADDRESS:
2545 // Repackage the long as a pointer.
2546 val = ConvL2X(val);
2547 val = _gvn.transform(new CastX2PNode(val));
2548 break;
2549 }
2550
2551 if (type == T_OBJECT) {
2552 store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type, mo, mismatched);
2553 } else {
2554 store_to_memory(control(), adr, val, type, adr_type, mo, requires_atomic_access, unaligned, mismatched);
2555 }
2556 }
2557
2558 switch(kind) {
2559 case Relaxed:
2560 case Opaque:
2561 case Release:
2562 break;
2563 case Acquire:
2564 case Volatile:
2565 if (!is_store) {
2566 insert_mem_bar(Op_MemBarAcquire);
2567 } else {
2568 if (!support_IRIW_for_not_multiple_copy_atomic_cpu) {
2569 insert_mem_bar(Op_MemBarVolatile);
2570 }
2571 }
2572 break;
2573 default:
2574 ShouldNotReachHere();
2575 }
2576
2577 if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);
2578
2579 return true;
2580 }
2581
2582 //----------------------------inline_unsafe_load_store----------------------------
2583 // This method serves a couple of different customers (depending on LoadStoreKind):
2584 //
2585 // LS_cmp_swap:
2586 //
2587 // boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
2588 // boolean compareAndSwapInt( Object o, long offset, int expected, int x);
2589 // boolean compareAndSwapLong( Object o, long offset, long expected, long x);
2590 //
2591 // LS_cmp_swap_weak:
2592 //
2593 // boolean weakCompareAndSwapObject( Object o, long offset, Object expected, Object x);
2594 // boolean weakCompareAndSwapObjectAcquire(Object o, long offset, Object expected, Object x);
2595 // boolean weakCompareAndSwapObjectRelease(Object o, long offset, Object expected, Object x);
2596 //
2597 // boolean weakCompareAndSwapInt( Object o, long offset, int expected, int x);
2598 // boolean weakCompareAndSwapIntAcquire( Object o, long offset, int expected, int x);
2599 // boolean weakCompareAndSwapIntRelease( Object o, long offset, int expected, int x);
2600 //
2601 // boolean weakCompareAndSwapLong( Object o, long offset, long expected, long x);
2602 // boolean weakCompareAndSwapLongAcquire( Object o, long offset, long expected, long x);
2603 // boolean weakCompareAndSwapLongRelease( Object o, long offset, long expected, long x);
2604 //
2605 // LS_cmp_exchange:
2606 //
2607 // Object compareAndExchangeObjectVolatile(Object o, long offset, Object expected, Object x);
2608 // Object compareAndExchangeObjectAcquire( Object o, long offset, Object expected, Object x);
2609 // Object compareAndExchangeObjectRelease( Object o, long offset, Object expected, Object x);
2610 //
2611 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2612 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2613 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2614 //
2615 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2616 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2617 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2618 //
2619 // LS_get_add:
2620 //
2621 // int getAndAddInt( Object o, long offset, int delta)
2622 // long getAndAddLong(Object o, long offset, long delta)
2623 //
2624 // LS_get_set:
2625 //
2626 // int getAndSet(Object o, long offset, int newValue)
2627 // long getAndSet(Object o, long offset, long newValue)
2628 // Object getAndSet(Object o, long offset, Object newValue)
2629 //
2630 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2631 // This basic scheme here is the same as inline_unsafe_access, but
2632 // differs in enough details that combining them would make the code
2633 // overly confusing. (This is a true fact! I originally combined
2634 // them, but even I was confused by it!) As much code/comments as
2635 // possible are retained from inline_unsafe_access though to make
2636 // the correspondences clearer. - dl
2637
2638 if (callee()->is_static()) return false; // caller must have the capability!
2639
2640 #ifndef PRODUCT
2641 BasicType rtype;
2642 {
2643 ResourceMark rm;
2644 // Check the signatures.
2645 ciSignature* sig = callee()->signature();
2646 rtype = sig->return_type()->basic_type();
2647 switch(kind) {
2648 case LS_get_add:
2649 case LS_get_set: {
2650 // Check the signatures.
2651 #ifdef ASSERT
2652 assert(rtype == type, "get and set must return the expected type");
2653 assert(sig->count() == 3, "get and set has 3 arguments");
2654 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2655 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2656 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2657 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2658 #endif // ASSERT
2659 break;
2660 }
2661 case LS_cmp_swap:
2662 case LS_cmp_swap_weak: {
2663 // Check the signatures.
2664 #ifdef ASSERT
2665 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2666 assert(sig->count() == 4, "CAS has 4 arguments");
2667 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2668 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2669 #endif // ASSERT
2670 break;
2671 }
2672 case LS_cmp_exchange: {
2673 // Check the signatures.
2674 #ifdef ASSERT
2675 assert(rtype == type, "CAS must return the expected type");
2676 assert(sig->count() == 4, "CAS has 4 arguments");
2677 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2678 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2679 #endif // ASSERT
2680 break;
2681 }
2682 default:
2683 ShouldNotReachHere();
2684 }
2685 }
2686 #endif //PRODUCT
2687
2688 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2689
2690 // Get arguments:
2691 Node* receiver = NULL;
2692 Node* base = NULL;
2693 Node* offset = NULL;
2694 Node* oldval = NULL;
2695 Node* newval = NULL;
2696 switch(kind) {
2697 case LS_cmp_swap:
2698 case LS_cmp_swap_weak:
2699 case LS_cmp_exchange: {
2700 const bool two_slot_type = type2size[type] == 2;
2701 receiver = argument(0); // type: oop
2702 base = argument(1); // type: oop
2703 offset = argument(2); // type: long
2704 oldval = argument(4); // type: oop, int, or long
2705 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2706 break;
2707 }
2708 case LS_get_add:
2709 case LS_get_set: {
2710 receiver = argument(0); // type: oop
2711 base = argument(1); // type: oop
2712 offset = argument(2); // type: long
2713 oldval = NULL;
2714 newval = argument(4); // type: oop, int, or long
2715 break;
2716 }
2717 default:
2718 ShouldNotReachHere();
2719 }
2720
2721 // Build field offset expression.
2722 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2723 // to be plain byte offsets, which are also the same as those accepted
2724 // by oopDesc::field_base.
2725 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2726 // 32-bit machines ignore the high half of long offsets
2727 offset = ConvL2X(offset);
2728 Node* adr = make_unsafe_address(base, offset);
2729 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2730
2731 Compile::AliasType* alias_type = C->alias_type(adr_type);
2732 BasicType bt = alias_type->basic_type();
2733 if (bt != T_ILLEGAL &&
2734 ((bt == T_OBJECT || bt == T_ARRAY) != (type == T_OBJECT))) {
2735 // Don't intrinsify mismatched object accesses.
2736 return false;
2737 }
2738
2739 // For CAS, unlike inline_unsafe_access, there seems no point in
2740 // trying to refine types. Just use the coarse types here.
2741 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2742 const Type *value_type = Type::get_const_basic_type(type);
2743
2744 switch (kind) {
2745 case LS_get_set:
2746 case LS_cmp_exchange: {
2747 if (type == T_OBJECT) {
2748 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2749 if (tjp != NULL) {
2750 value_type = tjp;
2751 }
2752 }
2753 break;
2754 }
2755 case LS_cmp_swap:
2756 case LS_cmp_swap_weak:
2757 case LS_get_add:
2758 break;
2759 default:
2760 ShouldNotReachHere();
2761 }
2762
2763 // Null check receiver.
2764 receiver = null_check(receiver);
2765 if (stopped()) {
2766 return true;
2767 }
2768
2769 int alias_idx = C->get_alias_index(adr_type);
2770
2771 // Memory-model-wise, a LoadStore acts like a little synchronized
2772 // block, so needs barriers on each side. These don't translate
2773 // into actual barriers on most machines, but we still need rest of
2774 // compiler to respect ordering.
2775
2776 switch (access_kind) {
2777 case Relaxed:
2778 case Acquire:
2779 break;
2780 case Release:
2781 insert_mem_bar(Op_MemBarRelease);
2782 break;
2783 case Volatile:
2784 if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
2785 insert_mem_bar(Op_MemBarVolatile);
2786 } else {
2787 insert_mem_bar(Op_MemBarRelease);
2788 }
2789 break;
2790 default:
2791 ShouldNotReachHere();
2792 }
2793 insert_mem_bar(Op_MemBarCPUOrder);
2794
2795 // Figure out the memory ordering.
2796 MemNode::MemOrd mo = access_kind_to_memord(access_kind);
2797
2798 // 4984716: MemBars must be inserted before this
2799 // memory node in order to avoid a false
2800 // dependency which will confuse the scheduler.
2801 Node *mem = memory(alias_idx);
2802
2803 // For now, we handle only those cases that actually exist: ints,
2804 // longs, and Object. Adding others should be straightforward.
2805 Node* load_store = NULL;
2806 switch(type) {
2807 case T_BYTE:
2808 switch(kind) {
2809 case LS_get_add:
2810 load_store = _gvn.transform(new GetAndAddBNode(control(), mem, adr, newval, adr_type));
2811 break;
2812 case LS_get_set:
2813 load_store = _gvn.transform(new GetAndSetBNode(control(), mem, adr, newval, adr_type));
2814 break;
2815 case LS_cmp_swap_weak:
2816 load_store = _gvn.transform(new WeakCompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2817 break;
2818 case LS_cmp_swap:
2819 load_store = _gvn.transform(new CompareAndSwapBNode(control(), mem, adr, newval, oldval, mo));
2820 break;
2821 case LS_cmp_exchange:
2822 load_store = _gvn.transform(new CompareAndExchangeBNode(control(), mem, adr, newval, oldval, adr_type, mo));
2823 break;
2824 default:
2825 ShouldNotReachHere();
2826 }
2827 break;
2828 case T_SHORT:
2829 switch(kind) {
2830 case LS_get_add:
2831 load_store = _gvn.transform(new GetAndAddSNode(control(), mem, adr, newval, adr_type));
2832 break;
2833 case LS_get_set:
2834 load_store = _gvn.transform(new GetAndSetSNode(control(), mem, adr, newval, adr_type));
2835 break;
2836 case LS_cmp_swap_weak:
2837 load_store = _gvn.transform(new WeakCompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2838 break;
2839 case LS_cmp_swap:
2840 load_store = _gvn.transform(new CompareAndSwapSNode(control(), mem, adr, newval, oldval, mo));
2841 break;
2842 case LS_cmp_exchange:
2843 load_store = _gvn.transform(new CompareAndExchangeSNode(control(), mem, adr, newval, oldval, adr_type, mo));
2844 break;
2845 default:
2846 ShouldNotReachHere();
2847 }
2848 break;
2849 case T_INT:
2850 switch(kind) {
2851 case LS_get_add:
2852 load_store = _gvn.transform(new GetAndAddINode(control(), mem, adr, newval, adr_type));
2853 break;
2854 case LS_get_set:
2855 load_store = _gvn.transform(new GetAndSetINode(control(), mem, adr, newval, adr_type));
2856 break;
2857 case LS_cmp_swap_weak:
2858 load_store = _gvn.transform(new WeakCompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2859 break;
2860 case LS_cmp_swap:
2861 load_store = _gvn.transform(new CompareAndSwapINode(control(), mem, adr, newval, oldval, mo));
2862 break;
2863 case LS_cmp_exchange:
2864 load_store = _gvn.transform(new CompareAndExchangeINode(control(), mem, adr, newval, oldval, adr_type, mo));
2865 break;
2866 default:
2867 ShouldNotReachHere();
2868 }
2869 break;
2870 case T_LONG:
2871 switch(kind) {
2872 case LS_get_add:
2873 load_store = _gvn.transform(new GetAndAddLNode(control(), mem, adr, newval, adr_type));
2874 break;
2875 case LS_get_set:
2876 load_store = _gvn.transform(new GetAndSetLNode(control(), mem, adr, newval, adr_type));
2877 break;
2878 case LS_cmp_swap_weak:
2879 load_store = _gvn.transform(new WeakCompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2880 break;
2881 case LS_cmp_swap:
2882 load_store = _gvn.transform(new CompareAndSwapLNode(control(), mem, adr, newval, oldval, mo));
2883 break;
2884 case LS_cmp_exchange:
2885 load_store = _gvn.transform(new CompareAndExchangeLNode(control(), mem, adr, newval, oldval, adr_type, mo));
2886 break;
2887 default:
2888 ShouldNotReachHere();
2889 }
2890 break;
2891 case T_OBJECT:
2892 // Transformation of a value which could be NULL pointer (CastPP #NULL)
2893 // could be delayed during Parse (for example, in adjust_map_after_if()).
2894 // Execute transformation here to avoid barrier generation in such case.
2895 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2896 newval = _gvn.makecon(TypePtr::NULL_PTR);
2897
2898 // Reference stores need a store barrier.
2899 switch(kind) {
2900 case LS_get_set: {
2901 // If pre-barrier must execute before the oop store, old value will require do_load here.
2902 if (!can_move_pre_barrier()) {
2903 pre_barrier(true /* do_load*/,
2904 control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
2905 NULL /* pre_val*/,
2906 T_OBJECT);
2907 } // Else move pre_barrier to use load_store value, see below.
2908 break;
2909 }
2910 case LS_cmp_swap_weak:
2911 case LS_cmp_swap:
2912 case LS_cmp_exchange: {
2913 // Same as for newval above:
2914 if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
2915 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2916 }
2917 // The only known value which might get overwritten is oldval.
2918 pre_barrier(false /* do_load */,
2919 control(), NULL, NULL, max_juint, NULL, NULL,
2920 oldval /* pre_val */,
2921 T_OBJECT);
2922 break;
2923 }
2924 default:
2925 ShouldNotReachHere();
2926 }
2927
2928 #ifdef _LP64
2929 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2930 Node *newval_enc = _gvn.transform(new EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2931
2932 switch(kind) {
2933 case LS_get_set:
2934 load_store = _gvn.transform(new GetAndSetNNode(control(), mem, adr, newval_enc, adr_type, value_type->make_narrowoop()));
2935 break;
2936 case LS_cmp_swap_weak: {
2937 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2938 load_store = _gvn.transform(new WeakCompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2939 break;
2940 }
2941 case LS_cmp_swap: {
2942 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2943 load_store = _gvn.transform(new CompareAndSwapNNode(control(), mem, adr, newval_enc, oldval_enc, mo));
2944 break;
2945 }
2946 case LS_cmp_exchange: {
2947 Node *oldval_enc = _gvn.transform(new EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2948 load_store = _gvn.transform(new CompareAndExchangeNNode(control(), mem, adr, newval_enc, oldval_enc, adr_type, value_type->make_narrowoop(), mo));
2949 break;
2950 }
2951 default:
2952 ShouldNotReachHere();
2953 }
2954 } else
2955 #endif
2956 switch (kind) {
2957 case LS_get_set:
2958 load_store = _gvn.transform(new GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2959 break;
2960 case LS_cmp_swap_weak:
2961 load_store = _gvn.transform(new WeakCompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2962 break;
2963 case LS_cmp_swap:
2964 load_store = _gvn.transform(new CompareAndSwapPNode(control(), mem, adr, newval, oldval, mo));
2965 break;
2966 case LS_cmp_exchange:
2967 load_store = _gvn.transform(new CompareAndExchangePNode(control(), mem, adr, newval, oldval, adr_type, value_type->is_oopptr(), mo));
2968 break;
2969 default:
2970 ShouldNotReachHere();
2971 }
2972
2973 // Emit the post barrier only when the actual store happened. This makes sense
2974 // to check only for LS_cmp_* that can fail to set the value.
2975 // LS_cmp_exchange does not produce any branches by default, so there is no
2976 // boolean result to piggyback on. TODO: When we merge CompareAndSwap with
2977 // CompareAndExchange and move branches here, it would make sense to conditionalize
2978 // post_barriers for LS_cmp_exchange as well.
2979 //
2980 // CAS success path is marked more likely since we anticipate this is a performance
2981 // critical path, while CAS failure path can use the penalty for going through unlikely
2982 // path as backoff. Which is still better than doing a store barrier there.
2983 switch (kind) {
2984 case LS_get_set:
2985 case LS_cmp_exchange: {
2986 post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2987 break;
2988 }
2989 case LS_cmp_swap_weak:
2990 case LS_cmp_swap: {
2991 IdealKit ideal(this);
2992 ideal.if_then(load_store, BoolTest::ne, ideal.ConI(0), PROB_STATIC_FREQUENT); {
2993 sync_kit(ideal);
2994 post_barrier(ideal.ctrl(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
2995 ideal.sync_kit(this);
2996 } ideal.end_if();
2997 final_sync(ideal);
2998 break;
2999 }
3000 default:
3001 ShouldNotReachHere();
3002 }
3003 break;
3004 default:
3005 fatal("unexpected type %d: %s", type, type2name(type));
3006 break;
3007 }
3008
3009 // SCMemProjNodes represent the memory state of a LoadStore. Their
3010 // main role is to prevent LoadStore nodes from being optimized away
3011 // when their results aren't used.
3012 Node* proj = _gvn.transform(new SCMemProjNode(load_store));
3013 set_memory(proj, alias_idx);
3014
3015 if (type == T_OBJECT && (kind == LS_get_set || kind == LS_cmp_exchange)) {
3016 #ifdef _LP64
3017 if (adr->bottom_type()->is_ptr_to_narrowoop()) {
3018 load_store = _gvn.transform(new DecodeNNode(load_store, load_store->get_ptr_type()));
3019 }
3020 #endif
3021 if (can_move_pre_barrier() && kind == LS_get_set) {
3022 // Don't need to load pre_val. The old value is returned by load_store.
3023 // The pre_barrier can execute after the xchg as long as no safepoint
3024 // gets inserted between them.
3025 pre_barrier(false /* do_load */,
3026 control(), NULL, NULL, max_juint, NULL, NULL,
3027 load_store /* pre_val */,
3028 T_OBJECT);
3029 }
3030 }
3031
3032 // Add the trailing membar surrounding the access
3033 insert_mem_bar(Op_MemBarCPUOrder);
3034
3035 switch (access_kind) {
3036 case Relaxed:
3037 case Release:
3038 break; // do nothing
3039 case Acquire:
3040 case Volatile:
3041 insert_mem_bar(Op_MemBarAcquire);
3042 // !support_IRIW_for_not_multiple_copy_atomic_cpu handled in platform code
3043 break;
3044 default:
3045 ShouldNotReachHere();
3046 }
3047
3048 assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3049 set_result(load_store);
3050 return true;
3051 }
3052
3053 MemNode::MemOrd LibraryCallKit::access_kind_to_memord_LS(AccessKind kind, bool is_store) {
3054 MemNode::MemOrd mo = MemNode::unset;
3055 switch(kind) {
3056 case Opaque:
3057 case Relaxed: mo = MemNode::unordered; break;
3058 case Acquire: mo = MemNode::acquire; break;
3059 case Release: mo = MemNode::release; break;
3060 case Volatile: mo = is_store ? MemNode::release : MemNode::acquire; break;
3061 default:
3062 ShouldNotReachHere();
3063 }
3064 guarantee(mo != MemNode::unset, "Should select memory ordering");
3065 return mo;
3066 }
3067
3068 MemNode::MemOrd LibraryCallKit::access_kind_to_memord(AccessKind kind) {
3069 MemNode::MemOrd mo = MemNode::unset;
3070 switch(kind) {
3071 case Opaque:
3072 case Relaxed: mo = MemNode::unordered; break;
3073 case Acquire: mo = MemNode::acquire; break;
3074 case Release: mo = MemNode::release; break;
3075 case Volatile: mo = MemNode::seqcst; break;
3076 default:
3077 ShouldNotReachHere();
3078 }
3079 guarantee(mo != MemNode::unset, "Should select memory ordering");
3080 return mo;
3081 }
3082
3083 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3084 // Regardless of form, don't allow previous ld/st to move down,
3085 // then issue acquire, release, or volatile mem_bar.
3086 insert_mem_bar(Op_MemBarCPUOrder);
3087 switch(id) {
3088 case vmIntrinsics::_loadFence:
3089 insert_mem_bar(Op_LoadFence);
3090 return true;
3091 case vmIntrinsics::_storeFence:
3092 insert_mem_bar(Op_StoreFence);
3093 return true;
3094 case vmIntrinsics::_fullFence:
3095 insert_mem_bar(Op_MemBarVolatile);
3096 return true;
3097 default:
3098 fatal_unexpected_iid(id);
3099 return false;
3100 }
3101 }
3102
3103 bool LibraryCallKit::inline_onspinwait() {
3104 insert_mem_bar(Op_OnSpinWait);
3105 return true;
3106 }
3107
3108 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3109 if (!kls->is_Con()) {
3110 return true;
3111 }
3112 const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
3113 if (klsptr == NULL) {
3114 return true;
3115 }
3116 ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
3117 // don't need a guard for a klass that is already initialized
3118 return !ik->is_initialized();
3119 }
3120
3121 //----------------------------inline_unsafe_allocate---------------------------
3122 // public native Object Unsafe.allocateInstance(Class<?> cls);
3123 bool LibraryCallKit::inline_unsafe_allocate() {
3124 if (callee()->is_static()) return false; // caller must have the capability!
3125
3126 null_check_receiver(); // null-check, then ignore
3127 Node* cls = null_check(argument(1));
3128 if (stopped()) return true;
3129
3130 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3131 kls = null_check(kls);
3132 if (stopped()) return true; // argument was like int.class
3133
3134 Node* test = NULL;
3135 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3136 // Note: The argument might still be an illegal value like
3137 // Serializable.class or Object[].class. The runtime will handle it.
3138 // But we must make an explicit check for initialization.
3139 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3140 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3141 // can generate code to load it as unsigned byte.
3142 Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3143 Node* bits = intcon(InstanceKlass::fully_initialized);
3144 test = _gvn.transform(new SubINode(inst, bits));
3145 // The 'test' is non-zero if we need to take a slow path.
3146 }
3147
3148 Node* obj = new_instance(kls, test);
3149 set_result(obj);
3150 return true;
3151 }
3152
3153 //------------------------inline_native_time_funcs--------------
3154 // inline code for System.currentTimeMillis() and System.nanoTime()
3155 // these have the same type and signature
3156 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3157 const TypeFunc* tf = OptoRuntime::void_long_Type();
3158 const TypePtr* no_memory_effects = NULL;
3159 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3160 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3161 #ifdef ASSERT
3162 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3163 assert(value_top == top(), "second value must be top");
3164 #endif
3165 set_result(value);
3166 return true;
3167 }
3168
3169 #ifdef TRACE_HAVE_INTRINSICS
3170
3171 /*
3172 * oop -> myklass
3173 * myklass->trace_id |= USED
3174 * return myklass->trace_id & ~0x3
3175 */
3176 bool LibraryCallKit::inline_native_classID() {
3177 Node* cls = null_check(argument(0), T_OBJECT);
3178 Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
3179 kls = null_check(kls, T_OBJECT);
3180
3181 ByteSize offset = TRACE_KLASS_TRACE_ID_OFFSET;
3182 Node* insp = basic_plus_adr(kls, in_bytes(offset));
3183 Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG, MemNode::unordered);
3184
3185 Node* clsused = longcon(0x01l); // set the class bit
3186 Node* orl = _gvn.transform(new OrLNode(tvalue, clsused));
3187 const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
3188 store_to_memory(control(), insp, orl, T_LONG, adr_type, MemNode::unordered);
3189
3190 #ifdef TRACE_ID_META_BITS
3191 Node* mbits = longcon(~TRACE_ID_META_BITS);
3192 tvalue = _gvn.transform(new AndLNode(tvalue, mbits));
3193 #endif
3194 #ifdef TRACE_ID_CLASS_SHIFT
3195 Node* cbits = intcon(TRACE_ID_CLASS_SHIFT);
3196 tvalue = _gvn.transform(new URShiftLNode(tvalue, cbits));
3197 #endif
3198
3199 set_result(tvalue);
3200 return true;
3201
3202 }
3203
3204 bool LibraryCallKit::inline_native_getBufferWriter() {
3205 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3206
3207 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
3208 in_bytes(TRACE_THREAD_DATA_WRITER_OFFSET)
3209 );
3210
3211 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3212
3213 Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
3214 Node* test_jobj_eq_null = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
3215
3216 IfNode* iff_jobj_null =
3217 create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
3218
3219 enum { _normal_path = 1,
3220 _null_path = 2,
3221 PATH_LIMIT };
3222
3223 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3224 PhiNode* result_val = new PhiNode(result_rgn, TypePtr::BOTTOM);
3225
3226 Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
3227 result_rgn->init_req(_null_path, jobj_is_null);
3228 result_val->init_req(_null_path, null());
3229
3230 Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
3231 result_rgn->init_req(_normal_path, jobj_is_not_null);
3232
3233 Node* res = make_load(jobj_is_not_null, jobj, TypeInstPtr::NOTNULL, T_OBJECT, MemNode::unordered);
3234 result_val->init_req(_normal_path, res);
3235
3236 set_result(result_rgn, result_val);
3237
3238 return true;
3239 }
3240
3241 #endif
3242
3243 //------------------------inline_native_currentThread------------------
3244 bool LibraryCallKit::inline_native_currentThread() {
3245 Node* junk = NULL;
3246 set_result(generate_current_thread(junk));
3247 return true;
3248 }
3249
3250 //------------------------inline_native_isInterrupted------------------
3251 // private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
3252 bool LibraryCallKit::inline_native_isInterrupted() {
3253 // Add a fast path to t.isInterrupted(clear_int):
3254 // (t == Thread.current() &&
3255 // (!TLS._osthread._interrupted || WINDOWS_ONLY(false) NOT_WINDOWS(!clear_int)))
3256 // ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
3257 // So, in the common case that the interrupt bit is false,
3258 // we avoid making a call into the VM. Even if the interrupt bit
3259 // is true, if the clear_int argument is false, we avoid the VM call.
3260 // However, if the receiver is not currentThread, we must call the VM,
3261 // because there must be some locking done around the operation.
3262
3263 // We only go to the fast case code if we pass two guards.
3264 // Paths which do not pass are accumulated in the slow_region.
3265
3266 enum {
3267 no_int_result_path = 1, // t == Thread.current() && !TLS._osthread._interrupted
3268 no_clear_result_path = 2, // t == Thread.current() && TLS._osthread._interrupted && !clear_int
3269 slow_result_path = 3, // slow path: t.isInterrupted(clear_int)
3270 PATH_LIMIT
3271 };
3272
3273 // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
3274 // out of the function.
3275 insert_mem_bar(Op_MemBarCPUOrder);
3276
3277 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3278 PhiNode* result_val = new PhiNode(result_rgn, TypeInt::BOOL);
3279
3280 RegionNode* slow_region = new RegionNode(1);
3281 record_for_igvn(slow_region);
3282
3283 // (a) Receiving thread must be the current thread.
3284 Node* rec_thr = argument(0);
3285 Node* tls_ptr = NULL;
3286 Node* cur_thr = generate_current_thread(tls_ptr);
3287 Node* cmp_thr = _gvn.transform(new CmpPNode(cur_thr, rec_thr));
3288 Node* bol_thr = _gvn.transform(new BoolNode(cmp_thr, BoolTest::ne));
3289
3290 generate_slow_guard(bol_thr, slow_region);
3291
3292 // (b) Interrupt bit on TLS must be false.
3293 Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
3294 Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3295 p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3296
3297 // Set the control input on the field _interrupted read to prevent it floating up.
3298 Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT, MemNode::unordered);
3299 Node* cmp_bit = _gvn.transform(new CmpINode(int_bit, intcon(0)));
3300 Node* bol_bit = _gvn.transform(new BoolNode(cmp_bit, BoolTest::ne));
3301
3302 IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
3303
3304 // First fast path: if (!TLS._interrupted) return false;
3305 Node* false_bit = _gvn.transform(new IfFalseNode(iff_bit));
3306 result_rgn->init_req(no_int_result_path, false_bit);
3307 result_val->init_req(no_int_result_path, intcon(0));
3308
3309 // drop through to next case
3310 set_control( _gvn.transform(new IfTrueNode(iff_bit)));
3311
3312 #ifndef _WINDOWS
3313 // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
3314 Node* clr_arg = argument(1);
3315 Node* cmp_arg = _gvn.transform(new CmpINode(clr_arg, intcon(0)));
3316 Node* bol_arg = _gvn.transform(new BoolNode(cmp_arg, BoolTest::ne));
3317 IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);
3318
3319 // Second fast path: ... else if (!clear_int) return true;
3320 Node* false_arg = _gvn.transform(new IfFalseNode(iff_arg));
3321 result_rgn->init_req(no_clear_result_path, false_arg);
3322 result_val->init_req(no_clear_result_path, intcon(1));
3323
3324 // drop through to next case
3325 set_control( _gvn.transform(new IfTrueNode(iff_arg)));
3326 #else
3327 // To return true on Windows you must read the _interrupted field
3328 // and check the event state i.e. take the slow path.
3329 #endif // _WINDOWS
3330
3331 // (d) Otherwise, go to the slow path.
3332 slow_region->add_req(control());
3333 set_control( _gvn.transform(slow_region));
3334
3335 if (stopped()) {
3336 // There is no slow path.
3337 result_rgn->init_req(slow_result_path, top());
3338 result_val->init_req(slow_result_path, top());
3339 } else {
3340 // non-virtual because it is a private non-static
3341 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);
3342
3343 Node* slow_val = set_results_for_java_call(slow_call);
3344 // this->control() comes from set_results_for_java_call
3345
3346 Node* fast_io = slow_call->in(TypeFunc::I_O);
3347 Node* fast_mem = slow_call->in(TypeFunc::Memory);
3348
3349 // These two phis are pre-filled with copies of of the fast IO and Memory
3350 PhiNode* result_mem = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
3351 PhiNode* result_io = PhiNode::make(result_rgn, fast_io, Type::ABIO);
3352
3353 result_rgn->init_req(slow_result_path, control());
3354 result_io ->init_req(slow_result_path, i_o());
3355 result_mem->init_req(slow_result_path, reset_memory());
3356 result_val->init_req(slow_result_path, slow_val);
3357
3358 set_all_memory(_gvn.transform(result_mem));
3359 set_i_o( _gvn.transform(result_io));
3360 }
3361
3362 C->set_has_split_ifs(true); // Has chance for split-if optimization
3363 set_result(result_rgn, result_val);
3364 return true;
3365 }
3366
3367 //---------------------------load_mirror_from_klass----------------------------
3368 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3369 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3370 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3371 return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT, MemNode::unordered);
3372 }
3373
3374 //-----------------------load_klass_from_mirror_common-------------------------
3375 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3376 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3377 // and branch to the given path on the region.
3378 // If never_see_null, take an uncommon trap on null, so we can optimistically
3379 // compile for the non-null case.
3380 // If the region is NULL, force never_see_null = true.
3381 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3382 bool never_see_null,
3383 RegionNode* region,
3384 int null_path,
3385 int offset) {
3386 if (region == NULL) never_see_null = true;
3387 Node* p = basic_plus_adr(mirror, offset);
3388 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3389 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3390 Node* null_ctl = top();
3391 kls = null_check_oop(kls, &null_ctl, never_see_null);
3392 if (region != NULL) {
3393 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3394 region->init_req(null_path, null_ctl);
3395 } else {
3396 assert(null_ctl == top(), "no loose ends");
3397 }
3398 return kls;
3399 }
3400
3401 //--------------------(inline_native_Class_query helpers)---------------------
3402 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3403 // Fall through if (mods & mask) == bits, take the guard otherwise.
3404 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3405 // Branch around if the given klass has the given modifier bit set.
3406 // Like generate_guard, adds a new path onto the region.
3407 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3408 Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3409 Node* mask = intcon(modifier_mask);
3410 Node* bits = intcon(modifier_bits);
3411 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3412 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3413 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3414 return generate_fair_guard(bol, region);
3415 }
3416 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3417 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3418 }
3419
3420 //-------------------------inline_native_Class_query-------------------
3421 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3422 const Type* return_type = TypeInt::BOOL;
3423 Node* prim_return_value = top(); // what happens if it's a primitive class?
3424 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3425 bool expect_prim = false; // most of these guys expect to work on refs
3426
3427 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3428
3429 Node* mirror = argument(0);
3430 Node* obj = top();
3431
3432 switch (id) {
3433 case vmIntrinsics::_isInstance:
3434 // nothing is an instance of a primitive type
3435 prim_return_value = intcon(0);
3436 obj = argument(1);
3437 break;
3438 case vmIntrinsics::_getModifiers:
3439 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3440 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3441 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3442 break;
3443 case vmIntrinsics::_isInterface:
3444 prim_return_value = intcon(0);
3445 break;
3446 case vmIntrinsics::_isArray:
3447 prim_return_value = intcon(0);
3448 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3449 break;
3450 case vmIntrinsics::_isPrimitive:
3451 prim_return_value = intcon(1);
3452 expect_prim = true; // obviously
3453 break;
3454 case vmIntrinsics::_getSuperclass:
3455 prim_return_value = null();
3456 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3457 break;
3458 case vmIntrinsics::_getClassAccessFlags:
3459 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3460 return_type = TypeInt::INT; // not bool! 6297094
3461 break;
3462 default:
3463 fatal_unexpected_iid(id);
3464 break;
3465 }
3466
3467 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3468 if (mirror_con == NULL) return false; // cannot happen?
3469
3470 #ifndef PRODUCT
3471 if (C->print_intrinsics() || C->print_inlining()) {
3472 ciType* k = mirror_con->java_mirror_type();
3473 if (k) {
3474 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3475 k->print_name();
3476 tty->cr();
3477 }
3478 }
3479 #endif
3480
3481 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3482 RegionNode* region = new RegionNode(PATH_LIMIT);
3483 record_for_igvn(region);
3484 PhiNode* phi = new PhiNode(region, return_type);
3485
3486 // The mirror will never be null of Reflection.getClassAccessFlags, however
3487 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3488 // if it is. See bug 4774291.
3489
3490 // For Reflection.getClassAccessFlags(), the null check occurs in
3491 // the wrong place; see inline_unsafe_access(), above, for a similar
3492 // situation.
3493 mirror = null_check(mirror);
3494 // If mirror or obj is dead, only null-path is taken.
3495 if (stopped()) return true;
3496
3497 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3498
3499 // Now load the mirror's klass metaobject, and null-check it.
3500 // Side-effects region with the control path if the klass is null.
3501 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3502 // If kls is null, we have a primitive mirror.
3503 phi->init_req(_prim_path, prim_return_value);
3504 if (stopped()) { set_result(region, phi); return true; }
3505 bool safe_for_replace = (region->in(_prim_path) == top());
3506
3507 Node* p; // handy temp
3508 Node* null_ctl;
3509
3510 // Now that we have the non-null klass, we can perform the real query.
3511 // For constant classes, the query will constant-fold in LoadNode::Value.
3512 Node* query_value = top();
3513 switch (id) {
3514 case vmIntrinsics::_isInstance:
3515 // nothing is an instance of a primitive type
3516 query_value = gen_instanceof(obj, kls, safe_for_replace);
3517 break;
3518
3519 case vmIntrinsics::_getModifiers:
3520 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3521 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3522 break;
3523
3524 case vmIntrinsics::_isInterface:
3525 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3526 if (generate_interface_guard(kls, region) != NULL)
3527 // A guard was added. If the guard is taken, it was an interface.
3528 phi->add_req(intcon(1));
3529 // If we fall through, it's a plain class.
3530 query_value = intcon(0);
3531 break;
3532
3533 case vmIntrinsics::_isArray:
3534 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3535 if (generate_array_guard(kls, region) != NULL)
3536 // A guard was added. If the guard is taken, it was an array.
3537 phi->add_req(intcon(1));
3538 // If we fall through, it's a plain class.
3539 query_value = intcon(0);
3540 break;
3541
3542 case vmIntrinsics::_isPrimitive:
3543 query_value = intcon(0); // "normal" path produces false
3544 break;
3545
3546 case vmIntrinsics::_getSuperclass:
3547 // The rules here are somewhat unfortunate, but we can still do better
3548 // with random logic than with a JNI call.
3549 // Interfaces store null or Object as _super, but must report null.
3550 // Arrays store an intermediate super as _super, but must report Object.
3551 // Other types can report the actual _super.
3552 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3553 if (generate_interface_guard(kls, region) != NULL)
3554 // A guard was added. If the guard is taken, it was an interface.
3555 phi->add_req(null());
3556 if (generate_array_guard(kls, region) != NULL)
3557 // A guard was added. If the guard is taken, it was an array.
3558 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3559 // If we fall through, it's a plain class. Get its _super.
3560 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3561 kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
3562 null_ctl = top();
3563 kls = null_check_oop(kls, &null_ctl);
3564 if (null_ctl != top()) {
3565 // If the guard is taken, Object.superClass is null (both klass and mirror).
3566 region->add_req(null_ctl);
3567 phi ->add_req(null());
3568 }
3569 if (!stopped()) {
3570 query_value = load_mirror_from_klass(kls);
3571 }
3572 break;
3573
3574 case vmIntrinsics::_getClassAccessFlags:
3575 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3576 query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3577 break;
3578
3579 default:
3580 fatal_unexpected_iid(id);
3581 break;
3582 }
3583
3584 // Fall-through is the normal case of a query to a real class.
3585 phi->init_req(1, query_value);
3586 region->init_req(1, control());
3587
3588 C->set_has_split_ifs(true); // Has chance for split-if optimization
3589 set_result(region, phi);
3590 return true;
3591 }
3592
3593 //-------------------------inline_Class_cast-------------------
3594 bool LibraryCallKit::inline_Class_cast() {
3595 Node* mirror = argument(0); // Class
3596 Node* obj = argument(1);
3597 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3598 if (mirror_con == NULL) {
3599 return false; // dead path (mirror->is_top()).
3600 }
3601 if (obj == NULL || obj->is_top()) {
3602 return false; // dead path
3603 }
3604 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3605
3606 // First, see if Class.cast() can be folded statically.
3607 // java_mirror_type() returns non-null for compile-time Class constants.
3608 ciType* tm = mirror_con->java_mirror_type();
3609 if (tm != NULL && tm->is_klass() &&
3610 tp != NULL && tp->klass() != NULL) {
3611 if (!tp->klass()->is_loaded()) {
3612 // Don't use intrinsic when class is not loaded.
3613 return false;
3614 } else {
3615 int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3616 if (static_res == Compile::SSC_always_true) {
3617 // isInstance() is true - fold the code.
3618 set_result(obj);
3619 return true;
3620 } else if (static_res == Compile::SSC_always_false) {
3621 // Don't use intrinsic, have to throw ClassCastException.
3622 // If the reference is null, the non-intrinsic bytecode will
3623 // be optimized appropriately.
3624 return false;
3625 }
3626 }
3627 }
3628
3629 // Bailout intrinsic and do normal inlining if exception path is frequent.
3630 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3631 return false;
3632 }
3633
3634 // Generate dynamic checks.
3635 // Class.cast() is java implementation of _checkcast bytecode.
3636 // Do checkcast (Parse::do_checkcast()) optimizations here.
3637
3638 mirror = null_check(mirror);
3639 // If mirror is dead, only null-path is taken.
3640 if (stopped()) {
3641 return true;
3642 }
3643
3644 // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3645 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3646 RegionNode* region = new RegionNode(PATH_LIMIT);
3647 record_for_igvn(region);
3648
3649 // Now load the mirror's klass metaobject, and null-check it.
3650 // If kls is null, we have a primitive mirror and
3651 // nothing is an instance of a primitive type.
3652 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3653
3654 Node* res = top();
3655 if (!stopped()) {
3656 Node* bad_type_ctrl = top();
3657 // Do checkcast optimizations.
3658 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3659 region->init_req(_bad_type_path, bad_type_ctrl);
3660 }
3661 if (region->in(_prim_path) != top() ||
3662 region->in(_bad_type_path) != top()) {
3663 // Let Interpreter throw ClassCastException.
3664 PreserveJVMState pjvms(this);
3665 set_control(_gvn.transform(region));
3666 uncommon_trap(Deoptimization::Reason_intrinsic,
3667 Deoptimization::Action_maybe_recompile);
3668 }
3669 if (!stopped()) {
3670 set_result(res);
3671 }
3672 return true;
3673 }
3674
3675
3676 //--------------------------inline_native_subtype_check------------------------
3677 // This intrinsic takes the JNI calls out of the heart of
3678 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3679 bool LibraryCallKit::inline_native_subtype_check() {
3680 // Pull both arguments off the stack.
3681 Node* args[2]; // two java.lang.Class mirrors: superc, subc
3682 args[0] = argument(0);
3683 args[1] = argument(1);
3684 Node* klasses[2]; // corresponding Klasses: superk, subk
3685 klasses[0] = klasses[1] = top();
3686
3687 enum {
3688 // A full decision tree on {superc is prim, subc is prim}:
3689 _prim_0_path = 1, // {P,N} => false
3690 // {P,P} & superc!=subc => false
3691 _prim_same_path, // {P,P} & superc==subc => true
3692 _prim_1_path, // {N,P} => false
3693 _ref_subtype_path, // {N,N} & subtype check wins => true
3694 _both_ref_path, // {N,N} & subtype check loses => false
3695 PATH_LIMIT
3696 };
3697
3698 RegionNode* region = new RegionNode(PATH_LIMIT);
3699 Node* phi = new PhiNode(region, TypeInt::BOOL);
3700 record_for_igvn(region);
3701
3702 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
3703 const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3704 int class_klass_offset = java_lang_Class::klass_offset_in_bytes();
3705
3706 // First null-check both mirrors and load each mirror's klass metaobject.
3707 int which_arg;
3708 for (which_arg = 0; which_arg <= 1; which_arg++) {
3709 Node* arg = args[which_arg];
3710 arg = null_check(arg);
3711 if (stopped()) break;
3712 args[which_arg] = arg;
3713
3714 Node* p = basic_plus_adr(arg, class_klass_offset);
3715 Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3716 klasses[which_arg] = _gvn.transform(kls);
3717 }
3718
3719 // Having loaded both klasses, test each for null.
3720 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3721 for (which_arg = 0; which_arg <= 1; which_arg++) {
3722 Node* kls = klasses[which_arg];
3723 Node* null_ctl = top();
3724 kls = null_check_oop(kls, &null_ctl, never_see_null);
3725 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3726 region->init_req(prim_path, null_ctl);
3727 if (stopped()) break;
3728 klasses[which_arg] = kls;
3729 }
3730
3731 if (!stopped()) {
3732 // now we have two reference types, in klasses[0..1]
3733 Node* subk = klasses[1]; // the argument to isAssignableFrom
3734 Node* superk = klasses[0]; // the receiver
3735 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3736 // now we have a successful reference subtype check
3737 region->set_req(_ref_subtype_path, control());
3738 }
3739
3740 // If both operands are primitive (both klasses null), then
3741 // we must return true when they are identical primitives.
3742 // It is convenient to test this after the first null klass check.
3743 set_control(region->in(_prim_0_path)); // go back to first null check
3744 if (!stopped()) {
3745 // Since superc is primitive, make a guard for the superc==subc case.
3746 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3747 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3748 generate_guard(bol_eq, region, PROB_FAIR);
3749 if (region->req() == PATH_LIMIT+1) {
3750 // A guard was added. If the added guard is taken, superc==subc.
3751 region->swap_edges(PATH_LIMIT, _prim_same_path);
3752 region->del_req(PATH_LIMIT);
3753 }
3754 region->set_req(_prim_0_path, control()); // Not equal after all.
3755 }
3756
3757 // these are the only paths that produce 'true':
3758 phi->set_req(_prim_same_path, intcon(1));
3759 phi->set_req(_ref_subtype_path, intcon(1));
3760
3761 // pull together the cases:
3762 assert(region->req() == PATH_LIMIT, "sane region");
3763 for (uint i = 1; i < region->req(); i++) {
3764 Node* ctl = region->in(i);
3765 if (ctl == NULL || ctl == top()) {
3766 region->set_req(i, top());
3767 phi ->set_req(i, top());
3768 } else if (phi->in(i) == NULL) {
3769 phi->set_req(i, intcon(0)); // all other paths produce 'false'
3770 }
3771 }
3772
3773 set_control(_gvn.transform(region));
3774 set_result(_gvn.transform(phi));
3775 return true;
3776 }
3777
3778 //---------------------generate_array_guard_common------------------------
3779 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3780 bool obj_array, bool not_array) {
3781
3782 if (stopped()) {
3783 return NULL;
3784 }
3785
3786 // If obj_array/non_array==false/false:
3787 // Branch around if the given klass is in fact an array (either obj or prim).
3788 // If obj_array/non_array==false/true:
3789 // Branch around if the given klass is not an array klass of any kind.
3790 // If obj_array/non_array==true/true:
3791 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3792 // If obj_array/non_array==true/false:
3793 // Branch around if the kls is an oop array (Object[] or subtype)
3794 //
3795 // Like generate_guard, adds a new path onto the region.
3796 jint layout_con = 0;
3797 Node* layout_val = get_layout_helper(kls, layout_con);
3798 if (layout_val == NULL) {
3799 bool query = (obj_array
3800 ? Klass::layout_helper_is_objArray(layout_con)
3801 : Klass::layout_helper_is_array(layout_con));
3802 if (query == not_array) {
3803 return NULL; // never a branch
3804 } else { // always a branch
3805 Node* always_branch = control();
3806 if (region != NULL)
3807 region->add_req(always_branch);
3808 set_control(top());
3809 return always_branch;
3810 }
3811 }
3812 // Now test the correct condition.
3813 jint nval = (obj_array
3814 ? (jint)(Klass::_lh_array_tag_type_value
3815 << Klass::_lh_array_tag_shift)
3816 : Klass::_lh_neutral_value);
3817 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3818 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
3819 // invert the test if we are looking for a non-array
3820 if (not_array) btest = BoolTest(btest).negate();
3821 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3822 return generate_fair_guard(bol, region);
3823 }
3824
3825
3826 //-----------------------inline_native_newArray--------------------------
3827 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3828 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3829 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3830 Node* mirror;
3831 Node* count_val;
3832 if (uninitialized) {
3833 mirror = argument(1);
3834 count_val = argument(2);
3835 } else {
3836 mirror = argument(0);
3837 count_val = argument(1);
3838 }
3839
3840 mirror = null_check(mirror);
3841 // If mirror or obj is dead, only null-path is taken.
3842 if (stopped()) return true;
3843
3844 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3845 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3846 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3847 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
3848 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3849
3850 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3851 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3852 result_reg, _slow_path);
3853 Node* normal_ctl = control();
3854 Node* no_array_ctl = result_reg->in(_slow_path);
3855
3856 // Generate code for the slow case. We make a call to newArray().
3857 set_control(no_array_ctl);
3858 if (!stopped()) {
3859 // Either the input type is void.class, or else the
3860 // array klass has not yet been cached. Either the
3861 // ensuing call will throw an exception, or else it
3862 // will cache the array klass for next time.
3863 PreserveJVMState pjvms(this);
3864 CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3865 Node* slow_result = set_results_for_java_call(slow_call);
3866 // this->control() comes from set_results_for_java_call
3867 result_reg->set_req(_slow_path, control());
3868 result_val->set_req(_slow_path, slow_result);
3869 result_io ->set_req(_slow_path, i_o());
3870 result_mem->set_req(_slow_path, reset_memory());
3871 }
3872
3873 set_control(normal_ctl);
3874 if (!stopped()) {
3875 // Normal case: The array type has been cached in the java.lang.Class.
3876 // The following call works fine even if the array type is polymorphic.
3877 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3878 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
3879 result_reg->init_req(_normal_path, control());
3880 result_val->init_req(_normal_path, obj);
3881 result_io ->init_req(_normal_path, i_o());
3882 result_mem->init_req(_normal_path, reset_memory());
3883
3884 if (uninitialized) {
3885 // Mark the allocation so that zeroing is skipped
3886 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3887 alloc->maybe_set_complete(&_gvn);
3888 }
3889 }
3890
3891 // Return the combined state.
3892 set_i_o( _gvn.transform(result_io) );
3893 set_all_memory( _gvn.transform(result_mem));
3894
3895 C->set_has_split_ifs(true); // Has chance for split-if optimization
3896 set_result(result_reg, result_val);
3897 return true;
3898 }
3899
3900 //----------------------inline_native_getLength--------------------------
3901 // public static native int java.lang.reflect.Array.getLength(Object array);
3902 bool LibraryCallKit::inline_native_getLength() {
3903 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3904
3905 Node* array = null_check(argument(0));
3906 // If array is dead, only null-path is taken.
3907 if (stopped()) return true;
3908
3909 // Deoptimize if it is a non-array.
3910 Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3911
3912 if (non_array != NULL) {
3913 PreserveJVMState pjvms(this);
3914 set_control(non_array);
3915 uncommon_trap(Deoptimization::Reason_intrinsic,
3916 Deoptimization::Action_maybe_recompile);
3917 }
3918
3919 // If control is dead, only non-array-path is taken.
3920 if (stopped()) return true;
3921
3922 // The works fine even if the array type is polymorphic.
3923 // It could be a dynamic mix of int[], boolean[], Object[], etc.
3924 Node* result = load_array_length(array);
3925
3926 C->set_has_split_ifs(true); // Has chance for split-if optimization
3927 set_result(result);
3928 return true;
3929 }
3930
3931 //------------------------inline_array_copyOf----------------------------
3932 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
3933 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
3934 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3935 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
3936
3937 // Get the arguments.
3938 Node* original = argument(0);
3939 Node* start = is_copyOfRange? argument(1): intcon(0);
3940 Node* end = is_copyOfRange? argument(2): argument(1);
3941 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3942
3943 Node* newcopy = NULL;
3944
3945 // Set the original stack and the reexecute bit for the interpreter to reexecute
3946 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3947 { PreserveReexecuteState preexecs(this);
3948 jvms()->set_should_reexecute(true);
3949
3950 array_type_mirror = null_check(array_type_mirror);
3951 original = null_check(original);
3952
3953 // Check if a null path was taken unconditionally.
3954 if (stopped()) return true;
3955
3956 Node* orig_length = load_array_length(original);
3957
3958 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3959 klass_node = null_check(klass_node);
3960
3961 RegionNode* bailout = new RegionNode(1);
3962 record_for_igvn(bailout);
3963
3964 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3965 // Bail out if that is so.
3966 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3967 if (not_objArray != NULL) {
3968 // Improve the klass node's type from the new optimistic assumption:
3969 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3970 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3971 Node* cast = new CastPPNode(klass_node, akls);
3972 cast->init_req(0, control());
3973 klass_node = _gvn.transform(cast);
3974 }
3975
3976 // Bail out if either start or end is negative.
3977 generate_negative_guard(start, bailout, &start);
3978 generate_negative_guard(end, bailout, &end);
3979
3980 Node* length = end;
3981 if (_gvn.type(start) != TypeInt::ZERO) {
3982 length = _gvn.transform(new SubINode(end, start));
3983 }
3984
3985 // Bail out if length is negative.
3986 // Without this the new_array would throw
3987 // NegativeArraySizeException but IllegalArgumentException is what
3988 // should be thrown
3989 generate_negative_guard(length, bailout, &length);
3990
3991 if (bailout->req() > 1) {
3992 PreserveJVMState pjvms(this);
3993 set_control(_gvn.transform(bailout));
3994 uncommon_trap(Deoptimization::Reason_intrinsic,
3995 Deoptimization::Action_maybe_recompile);
3996 }
3997
3998 if (!stopped()) {
3999 // How many elements will we copy from the original?
4000 // The answer is MinI(orig_length - start, length).
4001 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4002 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4003
4004 // Generate a direct call to the right arraycopy function(s).
4005 // We know the copy is disjoint but we might not know if the
4006 // oop stores need checking.
4007 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4008 // This will fail a store-check if x contains any non-nulls.
4009
4010 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4011 // loads/stores but it is legal only if we're sure the
4012 // Arrays.copyOf would succeed. So we need all input arguments
4013 // to the copyOf to be validated, including that the copy to the
4014 // new array won't trigger an ArrayStoreException. That subtype
4015 // check can be optimized if we know something on the type of
4016 // the input array from type speculation.
4017 if (_gvn.type(klass_node)->singleton()) {
4018 ciKlass* subk = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
4019 ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
4020
4021 int test = C->static_subtype_check(superk, subk);
4022 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4023 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4024 if (t_original->speculative_type() != NULL) {
4025 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4026 }
4027 }
4028 }
4029
4030 bool validated = false;
4031 // Reason_class_check rather than Reason_intrinsic because we
4032 // want to intrinsify even if this traps.
4033 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4034 Node* not_subtype_ctrl = gen_subtype_check(load_object_klass(original),
4035 klass_node);
4036
4037 if (not_subtype_ctrl != top()) {
4038 PreserveJVMState pjvms(this);
4039 set_control(not_subtype_ctrl);
4040 uncommon_trap(Deoptimization::Reason_class_check,
4041 Deoptimization::Action_make_not_entrant);
4042 assert(stopped(), "Should be stopped");
4043 }
4044 validated = true;
4045 }
4046
4047 if (!stopped()) {
4048 newcopy = new_array(klass_node, length, 0); // no arguments to push
4049
4050 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4051 load_object_klass(original), klass_node);
4052 if (!is_copyOfRange) {
4053 ac->set_copyof(validated);
4054 } else {
4055 ac->set_copyofrange(validated);
4056 }
4057 Node* n = _gvn.transform(ac);
4058 if (n == ac) {
4059 ac->connect_outputs(this);
4060 } else {
4061 assert(validated, "shouldn't transform if all arguments not validated");
4062 set_all_memory(n);
4063 }
4064 }
4065 }
4066 } // original reexecute is set back here
4067
4068 C->set_has_split_ifs(true); // Has chance for split-if optimization
4069 if (!stopped()) {
4070 set_result(newcopy);
4071 }
4072 return true;
4073 }
4074
4075
4076 //----------------------generate_virtual_guard---------------------------
4077 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4078 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4079 RegionNode* slow_region) {
4080 ciMethod* method = callee();
4081 int vtable_index = method->vtable_index();
4082 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4083 "bad index %d", vtable_index);
4084 // Get the Method* out of the appropriate vtable entry.
4085 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4086 vtable_index*vtableEntry::size_in_bytes() +
4087 vtableEntry::method_offset_in_bytes();
4088 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4089 Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4090
4091 // Compare the target method with the expected method (e.g., Object.hashCode).
4092 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4093
4094 Node* native_call = makecon(native_call_addr);
4095 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4096 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4097
4098 return generate_slow_guard(test_native, slow_region);
4099 }
4100
4101 //-----------------------generate_method_call----------------------------
4102 // Use generate_method_call to make a slow-call to the real
4103 // method if the fast path fails. An alternative would be to
4104 // use a stub like OptoRuntime::slow_arraycopy_Java.
4105 // This only works for expanding the current library call,
4106 // not another intrinsic. (E.g., don't use this for making an
4107 // arraycopy call inside of the copyOf intrinsic.)
4108 CallJavaNode*
4109 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4110 // When compiling the intrinsic method itself, do not use this technique.
4111 guarantee(callee() != C->method(), "cannot make slow-call to self");
4112
4113 ciMethod* method = callee();
4114 // ensure the JVMS we have will be correct for this call
4115 guarantee(method_id == method->intrinsic_id(), "must match");
4116
4117 const TypeFunc* tf = TypeFunc::make(method);
4118 CallJavaNode* slow_call;
4119 if (is_static) {
4120 assert(!is_virtual, "");
4121 slow_call = new CallStaticJavaNode(C, tf,
4122 SharedRuntime::get_resolve_static_call_stub(),
4123 method, bci());
4124 } else if (is_virtual) {
4125 null_check_receiver();
4126 int vtable_index = Method::invalid_vtable_index;
4127 if (UseInlineCaches) {
4128 // Suppress the vtable call
4129 } else {
4130 // hashCode and clone are not a miranda methods,
4131 // so the vtable index is fixed.
4132 // No need to use the linkResolver to get it.
4133 vtable_index = method->vtable_index();
4134 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4135 "bad index %d", vtable_index);
4136 }
4137 slow_call = new CallDynamicJavaNode(tf,
4138 SharedRuntime::get_resolve_virtual_call_stub(),
4139 method, vtable_index, bci());
4140 } else { // neither virtual nor static: opt_virtual
4141 null_check_receiver();
4142 slow_call = new CallStaticJavaNode(C, tf,
4143 SharedRuntime::get_resolve_opt_virtual_call_stub(),
4144 method, bci());
4145 slow_call->set_optimized_virtual(true);
4146 }
4147 set_arguments_for_java_call(slow_call);
4148 set_edges_for_java_call(slow_call);
4149 return slow_call;
4150 }
4151
4152
4153 /**
4154 * Build special case code for calls to hashCode on an object. This call may
4155 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4156 * slightly different code.
4157 */
4158 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4159 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4160 assert(!(is_virtual && is_static), "either virtual, special, or static");
4161
4162 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4163
4164 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4165 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4166 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4167 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4168 Node* obj = NULL;
4169 if (!is_static) {
4170 // Check for hashing null object
4171 obj = null_check_receiver();
4172 if (stopped()) return true; // unconditionally null
4173 result_reg->init_req(_null_path, top());
4174 result_val->init_req(_null_path, top());
4175 } else {
4176 // Do a null check, and return zero if null.
4177 // System.identityHashCode(null) == 0
4178 obj = argument(0);
4179 Node* null_ctl = top();
4180 obj = null_check_oop(obj, &null_ctl);
4181 result_reg->init_req(_null_path, null_ctl);
4182 result_val->init_req(_null_path, _gvn.intcon(0));
4183 }
4184
4185 // Unconditionally null? Then return right away.
4186 if (stopped()) {
4187 set_control( result_reg->in(_null_path));
4188 if (!stopped())
4189 set_result(result_val->in(_null_path));
4190 return true;
4191 }
4192
4193 // We only go to the fast case code if we pass a number of guards. The
4194 // paths which do not pass are accumulated in the slow_region.
4195 RegionNode* slow_region = new RegionNode(1);
4196 record_for_igvn(slow_region);
4197
4198 // If this is a virtual call, we generate a funny guard. We pull out
4199 // the vtable entry corresponding to hashCode() from the target object.
4200 // If the target method which we are calling happens to be the native
4201 // Object hashCode() method, we pass the guard. We do not need this
4202 // guard for non-virtual calls -- the caller is known to be the native
4203 // Object hashCode().
4204 if (is_virtual) {
4205 // After null check, get the object's klass.
4206 Node* obj_klass = load_object_klass(obj);
4207 generate_virtual_guard(obj_klass, slow_region);
4208 }
4209
4210 // Get the header out of the object, use LoadMarkNode when available
4211 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4212 // The control of the load must be NULL. Otherwise, the load can move before
4213 // the null check after castPP removal.
4214 Node* no_ctrl = NULL;
4215 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4216
4217 // Test the header to see if it is unlocked.
4218 Node *lock_mask = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4219 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4220 Node *unlocked_val = _gvn.MakeConX(markOopDesc::unlocked_value);
4221 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4222 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4223
4224 generate_slow_guard(test_unlocked, slow_region);
4225
4226 // Get the hash value and check to see that it has been properly assigned.
4227 // We depend on hash_mask being at most 32 bits and avoid the use of
4228 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4229 // vm: see markOop.hpp.
4230 Node *hash_mask = _gvn.intcon(markOopDesc::hash_mask);
4231 Node *hash_shift = _gvn.intcon(markOopDesc::hash_shift);
4232 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4233 // This hack lets the hash bits live anywhere in the mark object now, as long
4234 // as the shift drops the relevant bits into the low 32 bits. Note that
4235 // Java spec says that HashCode is an int so there's no point in capturing
4236 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4237 hshifted_header = ConvX2I(hshifted_header);
4238 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4239
4240 Node *no_hash_val = _gvn.intcon(markOopDesc::no_hash);
4241 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4242 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4243
4244 generate_slow_guard(test_assigned, slow_region);
4245
4246 Node* init_mem = reset_memory();
4247 // fill in the rest of the null path:
4248 result_io ->init_req(_null_path, i_o());
4249 result_mem->init_req(_null_path, init_mem);
4250
4251 result_val->init_req(_fast_path, hash_val);
4252 result_reg->init_req(_fast_path, control());
4253 result_io ->init_req(_fast_path, i_o());
4254 result_mem->init_req(_fast_path, init_mem);
4255
4256 // Generate code for the slow case. We make a call to hashCode().
4257 set_control(_gvn.transform(slow_region));
4258 if (!stopped()) {
4259 // No need for PreserveJVMState, because we're using up the present state.
4260 set_all_memory(init_mem);
4261 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4262 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4263 Node* slow_result = set_results_for_java_call(slow_call);
4264 // this->control() comes from set_results_for_java_call
4265 result_reg->init_req(_slow_path, control());
4266 result_val->init_req(_slow_path, slow_result);
4267 result_io ->set_req(_slow_path, i_o());
4268 result_mem ->set_req(_slow_path, reset_memory());
4269 }
4270
4271 // Return the combined state.
4272 set_i_o( _gvn.transform(result_io) );
4273 set_all_memory( _gvn.transform(result_mem));
4274
4275 set_result(result_reg, result_val);
4276 return true;
4277 }
4278
4279 //---------------------------inline_native_getClass----------------------------
4280 // public final native Class<?> java.lang.Object.getClass();
4281 //
4282 // Build special case code for calls to getClass on an object.
4283 bool LibraryCallKit::inline_native_getClass() {
4284 Node* obj = null_check_receiver();
4285 if (stopped()) return true;
4286 set_result(load_mirror_from_klass(load_object_klass(obj)));
4287 return true;
4288 }
4289
4290 //-----------------inline_native_Reflection_getCallerClass---------------------
4291 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4292 //
4293 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4294 //
4295 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4296 // in that it must skip particular security frames and checks for
4297 // caller sensitive methods.
4298 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4299 #ifndef PRODUCT
4300 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4301 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4302 }
4303 #endif
4304
4305 if (!jvms()->has_method()) {
4306 #ifndef PRODUCT
4307 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4308 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4309 }
4310 #endif
4311 return false;
4312 }
4313
4314 // Walk back up the JVM state to find the caller at the required
4315 // depth.
4316 JVMState* caller_jvms = jvms();
4317
4318 // Cf. JVM_GetCallerClass
4319 // NOTE: Start the loop at depth 1 because the current JVM state does
4320 // not include the Reflection.getCallerClass() frame.
4321 for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
4322 ciMethod* m = caller_jvms->method();
4323 switch (n) {
4324 case 0:
4325 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4326 break;
4327 case 1:
4328 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4329 if (!m->caller_sensitive()) {
4330 #ifndef PRODUCT
4331 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4332 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4333 }
4334 #endif
4335 return false; // bail-out; let JVM_GetCallerClass do the work
4336 }
4337 break;
4338 default:
4339 if (!m->is_ignored_by_security_stack_walk()) {
4340 // We have reached the desired frame; return the holder class.
4341 // Acquire method holder as java.lang.Class and push as constant.
4342 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4343 ciInstance* caller_mirror = caller_klass->java_mirror();
4344 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4345
4346 #ifndef PRODUCT
4347 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4348 tty->print_cr(" Succeeded: caller = %d) %s.%s, JVMS depth = %d", n, caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), jvms()->depth());
4349 tty->print_cr(" JVM state at this point:");
4350 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4351 ciMethod* m = jvms()->of_depth(i)->method();
4352 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4353 }
4354 }
4355 #endif
4356 return true;
4357 }
4358 break;
4359 }
4360 }
4361
4362 #ifndef PRODUCT
4363 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4364 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4365 tty->print_cr(" JVM state at this point:");
4366 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4367 ciMethod* m = jvms()->of_depth(i)->method();
4368 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4369 }
4370 }
4371 #endif
4372
4373 return false; // bail-out; let JVM_GetCallerClass do the work
4374 }
4375
4376 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4377 Node* arg = argument(0);
4378 Node* result = NULL;
4379
4380 switch (id) {
4381 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4382 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4383 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4384 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4385
4386 case vmIntrinsics::_doubleToLongBits: {
4387 // two paths (plus control) merge in a wood
4388 RegionNode *r = new RegionNode(3);
4389 Node *phi = new PhiNode(r, TypeLong::LONG);
4390
4391 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4392 // Build the boolean node
4393 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4394
4395 // Branch either way.
4396 // NaN case is less traveled, which makes all the difference.
4397 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4398 Node *opt_isnan = _gvn.transform(ifisnan);
4399 assert( opt_isnan->is_If(), "Expect an IfNode");
4400 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4401 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4402
4403 set_control(iftrue);
4404
4405 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4406 Node *slow_result = longcon(nan_bits); // return NaN
4407 phi->init_req(1, _gvn.transform( slow_result ));
4408 r->init_req(1, iftrue);
4409
4410 // Else fall through
4411 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4412 set_control(iffalse);
4413
4414 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4415 r->init_req(2, iffalse);
4416
4417 // Post merge
4418 set_control(_gvn.transform(r));
4419 record_for_igvn(r);
4420
4421 C->set_has_split_ifs(true); // Has chance for split-if optimization
4422 result = phi;
4423 assert(result->bottom_type()->isa_long(), "must be");
4424 break;
4425 }
4426
4427 case vmIntrinsics::_floatToIntBits: {
4428 // two paths (plus control) merge in a wood
4429 RegionNode *r = new RegionNode(3);
4430 Node *phi = new PhiNode(r, TypeInt::INT);
4431
4432 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4433 // Build the boolean node
4434 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4435
4436 // Branch either way.
4437 // NaN case is less traveled, which makes all the difference.
4438 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4439 Node *opt_isnan = _gvn.transform(ifisnan);
4440 assert( opt_isnan->is_If(), "Expect an IfNode");
4441 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4442 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4443
4444 set_control(iftrue);
4445
4446 static const jint nan_bits = 0x7fc00000;
4447 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4448 phi->init_req(1, _gvn.transform( slow_result ));
4449 r->init_req(1, iftrue);
4450
4451 // Else fall through
4452 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4453 set_control(iffalse);
4454
4455 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4456 r->init_req(2, iffalse);
4457
4458 // Post merge
4459 set_control(_gvn.transform(r));
4460 record_for_igvn(r);
4461
4462 C->set_has_split_ifs(true); // Has chance for split-if optimization
4463 result = phi;
4464 assert(result->bottom_type()->isa_int(), "must be");
4465 break;
4466 }
4467
4468 default:
4469 fatal_unexpected_iid(id);
4470 break;
4471 }
4472 set_result(_gvn.transform(result));
4473 return true;
4474 }
4475
4476 //----------------------inline_unsafe_copyMemory-------------------------
4477 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4478 bool LibraryCallKit::inline_unsafe_copyMemory() {
4479 if (callee()->is_static()) return false; // caller must have the capability!
4480 null_check_receiver(); // null-check receiver
4481 if (stopped()) return true;
4482
4483 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4484
4485 Node* src_ptr = argument(1); // type: oop
4486 Node* src_off = ConvL2X(argument(2)); // type: long
4487 Node* dst_ptr = argument(4); // type: oop
4488 Node* dst_off = ConvL2X(argument(5)); // type: long
4489 Node* size = ConvL2X(argument(7)); // type: long
4490
4491 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4492 "fieldOffset must be byte-scaled");
4493
4494 Node* src = make_unsafe_address(src_ptr, src_off);
4495 Node* dst = make_unsafe_address(dst_ptr, dst_off);
4496
4497 // Conservatively insert a memory barrier on all memory slices.
4498 // Do not let writes of the copy source or destination float below the copy.
4499 insert_mem_bar(Op_MemBarCPUOrder);
4500
4501 // Call it. Note that the length argument is not scaled.
4502 make_runtime_call(RC_LEAF|RC_NO_FP,
4503 OptoRuntime::fast_arraycopy_Type(),
4504 StubRoutines::unsafe_arraycopy(),
4505 "unsafe_arraycopy",
4506 TypeRawPtr::BOTTOM,
4507 src, dst, size XTOP);
4508
4509 // Do not let reads of the copy destination float above the copy.
4510 insert_mem_bar(Op_MemBarCPUOrder);
4511
4512 return true;
4513 }
4514
4515 //------------------------clone_coping-----------------------------------
4516 // Helper function for inline_native_clone.
4517 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
4518 assert(obj_size != NULL, "");
4519 Node* raw_obj = alloc_obj->in(1);
4520 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4521
4522 AllocateNode* alloc = NULL;
4523 if (ReduceBulkZeroing) {
4524 // We will be completely responsible for initializing this object -
4525 // mark Initialize node as complete.
4526 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4527 // The object was just allocated - there should be no any stores!
4528 guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4529 // Mark as complete_with_arraycopy so that on AllocateNode
4530 // expansion, we know this AllocateNode is initialized by an array
4531 // copy and a StoreStore barrier exists after the array copy.
4532 alloc->initialization()->set_complete_with_arraycopy();
4533 }
4534
4535 // Copy the fastest available way.
4536 // TODO: generate fields copies for small objects instead.
4537 Node* src = obj;
4538 Node* dest = alloc_obj;
4539 Node* size = _gvn.transform(obj_size);
4540
4541 // Exclude the header but include array length to copy by 8 bytes words.
4542 // Can't use base_offset_in_bytes(bt) since basic type is unknown.
4543 int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
4544 instanceOopDesc::base_offset_in_bytes();
4545 // base_off:
4546 // 8 - 32-bit VM
4547 // 12 - 64-bit VM, compressed klass
4548 // 16 - 64-bit VM, normal klass
4549 if (base_off % BytesPerLong != 0) {
4550 assert(UseCompressedClassPointers, "");
4551 if (is_array) {
4552 // Exclude length to copy by 8 bytes words.
4553 base_off += sizeof(int);
4554 } else {
4555 // Include klass to copy by 8 bytes words.
4556 base_off = instanceOopDesc::klass_offset_in_bytes();
4557 }
4558 assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
4559 }
4560 src = basic_plus_adr(src, base_off);
4561 dest = basic_plus_adr(dest, base_off);
4562
4563 // Compute the length also, if needed:
4564 Node* countx = size;
4565 countx = _gvn.transform(new SubXNode(countx, MakeConX(base_off)));
4566 countx = _gvn.transform(new URShiftXNode(countx, intcon(LogBytesPerLong) ));
4567
4568 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4569
4570 ArrayCopyNode* ac = ArrayCopyNode::make(this, false, src, NULL, dest, NULL, countx, false, false);
4571 ac->set_clonebasic();
4572 Node* n = _gvn.transform(ac);
4573 if (n == ac) {
4574 set_predefined_output_for_runtime_call(ac, ac->in(TypeFunc::Memory), raw_adr_type);
4575 } else {
4576 set_all_memory(n);
4577 }
4578
4579 // If necessary, emit some card marks afterwards. (Non-arrays only.)
4580 if (card_mark) {
4581 assert(!is_array, "");
4582 // Put in store barrier for any and all oops we are sticking
4583 // into this object. (We could avoid this if we could prove
4584 // that the object type contains no oop fields at all.)
4585 Node* no_particular_value = NULL;
4586 Node* no_particular_field = NULL;
4587 int raw_adr_idx = Compile::AliasIdxRaw;
4588 post_barrier(control(),
4589 memory(raw_adr_type),
4590 alloc_obj,
4591 no_particular_field,
4592 raw_adr_idx,
4593 no_particular_value,
4594 T_OBJECT,
4595 false);
4596 }
4597
4598 // Do not let reads from the cloned object float above the arraycopy.
4599 if (alloc != NULL) {
4600 // Do not let stores that initialize this object be reordered with
4601 // a subsequent store that would make this object accessible by
4602 // other threads.
4603 // Record what AllocateNode this StoreStore protects so that
4604 // escape analysis can go from the MemBarStoreStoreNode to the
4605 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4606 // based on the escape status of the AllocateNode.
4607 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
4608 } else {
4609 insert_mem_bar(Op_MemBarCPUOrder);
4610 }
4611 }
4612
4613 //------------------------inline_native_clone----------------------------
4614 // protected native Object java.lang.Object.clone();
4615 //
4616 // Here are the simple edge cases:
4617 // null receiver => normal trap
4618 // virtual and clone was overridden => slow path to out-of-line clone
4619 // not cloneable or finalizer => slow path to out-of-line Object.clone
4620 //
4621 // The general case has two steps, allocation and copying.
4622 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4623 //
4624 // Copying also has two cases, oop arrays and everything else.
4625 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4626 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4627 //
4628 // These steps fold up nicely if and when the cloned object's klass
4629 // can be sharply typed as an object array, a type array, or an instance.
4630 //
4631 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4632 PhiNode* result_val;
4633
4634 // Set the reexecute bit for the interpreter to reexecute
4635 // the bytecode that invokes Object.clone if deoptimization happens.
4636 { PreserveReexecuteState preexecs(this);
4637 jvms()->set_should_reexecute(true);
4638
4639 Node* obj = null_check_receiver();
4640 if (stopped()) return true;
4641
4642 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4643
4644 // If we are going to clone an instance, we need its exact type to
4645 // know the number and types of fields to convert the clone to
4646 // loads/stores. Maybe a speculative type can help us.
4647 if (!obj_type->klass_is_exact() &&
4648 obj_type->speculative_type() != NULL &&
4649 obj_type->speculative_type()->is_instance_klass()) {
4650 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4651 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4652 !spec_ik->has_injected_fields()) {
4653 ciKlass* k = obj_type->klass();
4654 if (!k->is_instance_klass() ||
4655 k->as_instance_klass()->is_interface() ||
4656 k->as_instance_klass()->has_subklass()) {
4657 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4658 }
4659 }
4660 }
4661
4662 Node* obj_klass = load_object_klass(obj);
4663 const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
4664 const TypeOopPtr* toop = ((tklass != NULL)
4665 ? tklass->as_instance_type()
4666 : TypeInstPtr::NOTNULL);
4667
4668 // Conservatively insert a memory barrier on all memory slices.
4669 // Do not let writes into the original float below the clone.
4670 insert_mem_bar(Op_MemBarCPUOrder);
4671
4672 // paths into result_reg:
4673 enum {
4674 _slow_path = 1, // out-of-line call to clone method (virtual or not)
4675 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
4676 _array_path, // plain array allocation, plus arrayof_long_arraycopy
4677 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
4678 PATH_LIMIT
4679 };
4680 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4681 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4682 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
4683 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4684 record_for_igvn(result_reg);
4685
4686 const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
4687 int raw_adr_idx = Compile::AliasIdxRaw;
4688
4689 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4690 if (array_ctl != NULL) {
4691 // It's an array.
4692 PreserveJVMState pjvms(this);
4693 set_control(array_ctl);
4694 Node* obj_length = load_array_length(obj);
4695 Node* obj_size = NULL;
4696 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size); // no arguments to push
4697
4698 if (!use_ReduceInitialCardMarks()) {
4699 // If it is an oop array, it requires very special treatment,
4700 // because card marking is required on each card of the array.
4701 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4702 if (is_obja != NULL) {
4703 PreserveJVMState pjvms2(this);
4704 set_control(is_obja);
4705 // Generate a direct call to the right arraycopy function(s).
4706 Node* alloc = tightly_coupled_allocation(alloc_obj, NULL);
4707 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, alloc != NULL, false);
4708 ac->set_cloneoop();
4709 Node* n = _gvn.transform(ac);
4710 assert(n == ac, "cannot disappear");
4711 ac->connect_outputs(this);
4712
4713 result_reg->init_req(_objArray_path, control());
4714 result_val->init_req(_objArray_path, alloc_obj);
4715 result_i_o ->set_req(_objArray_path, i_o());
4716 result_mem ->set_req(_objArray_path, reset_memory());
4717 }
4718 }
4719 // Otherwise, there are no card marks to worry about.
4720 // (We can dispense with card marks if we know the allocation
4721 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
4722 // causes the non-eden paths to take compensating steps to
4723 // simulate a fresh allocation, so that no further
4724 // card marks are required in compiled code to initialize
4725 // the object.)
4726
4727 if (!stopped()) {
4728 copy_to_clone(obj, alloc_obj, obj_size, true, false);
4729
4730 // Present the results of the copy.
4731 result_reg->init_req(_array_path, control());
4732 result_val->init_req(_array_path, alloc_obj);
4733 result_i_o ->set_req(_array_path, i_o());
4734 result_mem ->set_req(_array_path, reset_memory());
4735 }
4736 }
4737
4738 // We only go to the instance fast case code if we pass a number of guards.
4739 // The paths which do not pass are accumulated in the slow_region.
4740 RegionNode* slow_region = new RegionNode(1);
4741 record_for_igvn(slow_region);
4742 if (!stopped()) {
4743 // It's an instance (we did array above). Make the slow-path tests.
4744 // If this is a virtual call, we generate a funny guard. We grab
4745 // the vtable entry corresponding to clone() from the target object.
4746 // If the target method which we are calling happens to be the
4747 // Object clone() method, we pass the guard. We do not need this
4748 // guard for non-virtual calls; the caller is known to be the native
4749 // Object clone().
4750 if (is_virtual) {
4751 generate_virtual_guard(obj_klass, slow_region);
4752 }
4753
4754 // The object must be easily cloneable and must not have a finalizer.
4755 // Both of these conditions may be checked in a single test.
4756 // We could optimize the test further, but we don't care.
4757 generate_access_flags_guard(obj_klass,
4758 // Test both conditions:
4759 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4760 // Must be cloneable but not finalizer:
4761 JVM_ACC_IS_CLONEABLE_FAST,
4762 slow_region);
4763 }
4764
4765 if (!stopped()) {
4766 // It's an instance, and it passed the slow-path tests.
4767 PreserveJVMState pjvms(this);
4768 Node* obj_size = NULL;
4769 // Need to deoptimize on exception from allocation since Object.clone intrinsic
4770 // is reexecuted if deoptimization occurs and there could be problems when merging
4771 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4772 Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4773
4774 copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());
4775
4776 // Present the results of the slow call.
4777 result_reg->init_req(_instance_path, control());
4778 result_val->init_req(_instance_path, alloc_obj);
4779 result_i_o ->set_req(_instance_path, i_o());
4780 result_mem ->set_req(_instance_path, reset_memory());
4781 }
4782
4783 // Generate code for the slow case. We make a call to clone().
4784 set_control(_gvn.transform(slow_region));
4785 if (!stopped()) {
4786 PreserveJVMState pjvms(this);
4787 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4788 Node* slow_result = set_results_for_java_call(slow_call);
4789 // this->control() comes from set_results_for_java_call
4790 result_reg->init_req(_slow_path, control());
4791 result_val->init_req(_slow_path, slow_result);
4792 result_i_o ->set_req(_slow_path, i_o());
4793 result_mem ->set_req(_slow_path, reset_memory());
4794 }
4795
4796 // Return the combined state.
4797 set_control( _gvn.transform(result_reg));
4798 set_i_o( _gvn.transform(result_i_o));
4799 set_all_memory( _gvn.transform(result_mem));
4800 } // original reexecute is set back here
4801
4802 set_result(_gvn.transform(result_val));
4803 return true;
4804 }
4805
4806 // If we have a tighly coupled allocation, the arraycopy may take care
4807 // of the array initialization. If one of the guards we insert between
4808 // the allocation and the arraycopy causes a deoptimization, an
4809 // unitialized array will escape the compiled method. To prevent that
4810 // we set the JVM state for uncommon traps between the allocation and
4811 // the arraycopy to the state before the allocation so, in case of
4812 // deoptimization, we'll reexecute the allocation and the
4813 // initialization.
4814 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4815 if (alloc != NULL) {
4816 ciMethod* trap_method = alloc->jvms()->method();
4817 int trap_bci = alloc->jvms()->bci();
4818
4819 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &
4820 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4821 // Make sure there's no store between the allocation and the
4822 // arraycopy otherwise visible side effects could be rexecuted
4823 // in case of deoptimization and cause incorrect execution.
4824 bool no_interfering_store = true;
4825 Node* mem = alloc->in(TypeFunc::Memory);
4826 if (mem->is_MergeMem()) {
4827 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4828 Node* n = mms.memory();
4829 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4830 assert(n->is_Store(), "what else?");
4831 no_interfering_store = false;
4832 break;
4833 }
4834 }
4835 } else {
4836 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4837 Node* n = mms.memory();
4838 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4839 assert(n->is_Store(), "what else?");
4840 no_interfering_store = false;
4841 break;
4842 }
4843 }
4844 }
4845
4846 if (no_interfering_store) {
4847 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4848 uint size = alloc->req();
4849 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4850 old_jvms->set_map(sfpt);
4851 for (uint i = 0; i < size; i++) {
4852 sfpt->init_req(i, alloc->in(i));
4853 }
4854 // re-push array length for deoptimization
4855 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4856 old_jvms->set_sp(old_jvms->sp()+1);
4857 old_jvms->set_monoff(old_jvms->monoff()+1);
4858 old_jvms->set_scloff(old_jvms->scloff()+1);
4859 old_jvms->set_endoff(old_jvms->endoff()+1);
4860 old_jvms->set_should_reexecute(true);
4861
4862 sfpt->set_i_o(map()->i_o());
4863 sfpt->set_memory(map()->memory());
4864 sfpt->set_control(map()->control());
4865
4866 JVMState* saved_jvms = jvms();
4867 saved_reexecute_sp = _reexecute_sp;
4868
4869 set_jvms(sfpt->jvms());
4870 _reexecute_sp = jvms()->sp();
4871
4872 return saved_jvms;
4873 }
4874 }
4875 }
4876 return NULL;
4877 }
4878
4879 // In case of a deoptimization, we restart execution at the
4880 // allocation, allocating a new array. We would leave an uninitialized
4881 // array in the heap that GCs wouldn't expect. Move the allocation
4882 // after the traps so we don't allocate the array if we
4883 // deoptimize. This is possible because tightly_coupled_allocation()
4884 // guarantees there's no observer of the allocated array at this point
4885 // and the control flow is simple enough.
4886 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4887 int saved_reexecute_sp, uint new_idx) {
4888 if (saved_jvms != NULL && !stopped()) {
4889 assert(alloc != NULL, "only with a tightly coupled allocation");
4890 // restore JVM state to the state at the arraycopy
4891 saved_jvms->map()->set_control(map()->control());
4892 assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4893 assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4894 // If we've improved the types of some nodes (null check) while
4895 // emitting the guards, propagate them to the current state
4896 map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4897 set_jvms(saved_jvms);
4898 _reexecute_sp = saved_reexecute_sp;
4899
4900 // Remove the allocation from above the guards
4901 CallProjections callprojs;
4902 alloc->extract_projections(&callprojs, true);
4903 InitializeNode* init = alloc->initialization();
4904 Node* alloc_mem = alloc->in(TypeFunc::Memory);
4905 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4906 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4907 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4908
4909 // move the allocation here (after the guards)
4910 _gvn.hash_delete(alloc);
4911 alloc->set_req(TypeFunc::Control, control());
4912 alloc->set_req(TypeFunc::I_O, i_o());
4913 Node *mem = reset_memory();
4914 set_all_memory(mem);
4915 alloc->set_req(TypeFunc::Memory, mem);
4916 set_control(init->proj_out(TypeFunc::Control));
4917 set_i_o(callprojs.fallthrough_ioproj);
4918
4919 // Update memory as done in GraphKit::set_output_for_allocation()
4920 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4921 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4922 if (ary_type->isa_aryptr() && length_type != NULL) {
4923 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4924 }
4925 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4926 int elemidx = C->get_alias_index(telemref);
4927 set_memory(init->proj_out(TypeFunc::Memory), Compile::AliasIdxRaw);
4928 set_memory(init->proj_out(TypeFunc::Memory), elemidx);
4929
4930 Node* allocx = _gvn.transform(alloc);
4931 assert(allocx == alloc, "where has the allocation gone?");
4932 assert(dest->is_CheckCastPP(), "not an allocation result?");
4933
4934 _gvn.hash_delete(dest);
4935 dest->set_req(0, control());
4936 Node* destx = _gvn.transform(dest);
4937 assert(destx == dest, "where has the allocation result gone?");
4938 }
4939 }
4940
4941
4942 //------------------------------inline_arraycopy-----------------------
4943 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
4944 // Object dest, int destPos,
4945 // int length);
4946 bool LibraryCallKit::inline_arraycopy() {
4947 // Get the arguments.
4948 Node* src = argument(0); // type: oop
4949 Node* src_offset = argument(1); // type: int
4950 Node* dest = argument(2); // type: oop
4951 Node* dest_offset = argument(3); // type: int
4952 Node* length = argument(4); // type: int
4953
4954 uint new_idx = C->unique();
4955
4956 // Check for allocation before we add nodes that would confuse
4957 // tightly_coupled_allocation()
4958 AllocateArrayNode* alloc = tightly_coupled_allocation(dest, NULL);
4959
4960 int saved_reexecute_sp = -1;
4961 JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4962 // See arraycopy_restore_alloc_state() comment
4963 // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4964 // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4965 // if saved_jvms == NULL and alloc != NULL, we can’t emit any guards
4966 bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4967
4968 // The following tests must be performed
4969 // (1) src and dest are arrays.
4970 // (2) src and dest arrays must have elements of the same BasicType
4971 // (3) src and dest must not be null.
4972 // (4) src_offset must not be negative.
4973 // (5) dest_offset must not be negative.
4974 // (6) length must not be negative.
4975 // (7) src_offset + length must not exceed length of src.
4976 // (8) dest_offset + length must not exceed length of dest.
4977 // (9) each element of an oop array must be assignable
4978
4979 // (3) src and dest must not be null.
4980 // always do this here because we need the JVM state for uncommon traps
4981 Node* null_ctl = top();
4982 src = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
4983 assert(null_ctl->is_top(), "no null control here");
4984 dest = null_check(dest, T_ARRAY);
4985
4986 if (!can_emit_guards) {
4987 // if saved_jvms == NULL and alloc != NULL, we don't emit any
4988 // guards but the arraycopy node could still take advantage of a
4989 // tightly allocated allocation. tightly_coupled_allocation() is
4990 // called again to make sure it takes the null check above into
4991 // account: the null check is mandatory and if it caused an
4992 // uncommon trap to be emitted then the allocation can't be
4993 // considered tightly coupled in this context.
4994 alloc = tightly_coupled_allocation(dest, NULL);
4995 }
4996
4997 bool validated = false;
4998
4999 const Type* src_type = _gvn.type(src);
5000 const Type* dest_type = _gvn.type(dest);
5001 const TypeAryPtr* top_src = src_type->isa_aryptr();
5002 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5003
5004 // Do we have the type of src?
5005 bool has_src = (top_src != NULL && top_src->klass() != NULL);
5006 // Do we have the type of dest?
5007 bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5008 // Is the type for src from speculation?
5009 bool src_spec = false;
5010 // Is the type for dest from speculation?
5011 bool dest_spec = false;
5012
5013 if ((!has_src || !has_dest) && can_emit_guards) {
5014 // We don't have sufficient type information, let's see if
5015 // speculative types can help. We need to have types for both src
5016 // and dest so that it pays off.
5017
5018 // Do we already have or could we have type information for src
5019 bool could_have_src = has_src;
5020 // Do we already have or could we have type information for dest
5021 bool could_have_dest = has_dest;
5022
5023 ciKlass* src_k = NULL;
5024 if (!has_src) {
5025 src_k = src_type->speculative_type_not_null();
5026 if (src_k != NULL && src_k->is_array_klass()) {
5027 could_have_src = true;
5028 }
5029 }
5030
5031 ciKlass* dest_k = NULL;
5032 if (!has_dest) {
5033 dest_k = dest_type->speculative_type_not_null();
5034 if (dest_k != NULL && dest_k->is_array_klass()) {
5035 could_have_dest = true;
5036 }
5037 }
5038
5039 if (could_have_src && could_have_dest) {
5040 // This is going to pay off so emit the required guards
5041 if (!has_src) {
5042 src = maybe_cast_profiled_obj(src, src_k, true);
5043 src_type = _gvn.type(src);
5044 top_src = src_type->isa_aryptr();
5045 has_src = (top_src != NULL && top_src->klass() != NULL);
5046 src_spec = true;
5047 }
5048 if (!has_dest) {
5049 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5050 dest_type = _gvn.type(dest);
5051 top_dest = dest_type->isa_aryptr();
5052 has_dest = (top_dest != NULL && top_dest->klass() != NULL);
5053 dest_spec = true;
5054 }
5055 }
5056 }
5057
5058 if (has_src && has_dest && can_emit_guards) {
5059 BasicType src_elem = top_src->klass()->as_array_klass()->element_type()->basic_type();
5060 BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
5061 if (src_elem == T_ARRAY) src_elem = T_OBJECT;
5062 if (dest_elem == T_ARRAY) dest_elem = T_OBJECT;
5063
5064 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5065 // If both arrays are object arrays then having the exact types
5066 // for both will remove the need for a subtype check at runtime
5067 // before the call and may make it possible to pick a faster copy
5068 // routine (without a subtype check on every element)
5069 // Do we have the exact type of src?
5070 bool could_have_src = src_spec;
5071 // Do we have the exact type of dest?
5072 bool could_have_dest = dest_spec;
5073 ciKlass* src_k = top_src->klass();
5074 ciKlass* dest_k = top_dest->klass();
5075 if (!src_spec) {
5076 src_k = src_type->speculative_type_not_null();
5077 if (src_k != NULL && src_k->is_array_klass()) {
5078 could_have_src = true;
5079 }
5080 }
5081 if (!dest_spec) {
5082 dest_k = dest_type->speculative_type_not_null();
5083 if (dest_k != NULL && dest_k->is_array_klass()) {
5084 could_have_dest = true;
5085 }
5086 }
5087 if (could_have_src && could_have_dest) {
5088 // If we can have both exact types, emit the missing guards
5089 if (could_have_src && !src_spec) {
5090 src = maybe_cast_profiled_obj(src, src_k, true);
5091 }
5092 if (could_have_dest && !dest_spec) {
5093 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5094 }
5095 }
5096 }
5097 }
5098
5099 ciMethod* trap_method = method();
5100 int trap_bci = bci();
5101 if (saved_jvms != NULL) {
5102 trap_method = alloc->jvms()->method();
5103 trap_bci = alloc->jvms()->bci();
5104 }
5105
5106 bool negative_length_guard_generated = false;
5107
5108 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5109 can_emit_guards &&
5110 !src->is_top() && !dest->is_top()) {
5111 // validate arguments: enables transformation the ArrayCopyNode
5112 validated = true;
5113
5114 RegionNode* slow_region = new RegionNode(1);
5115 record_for_igvn(slow_region);
5116
5117 // (1) src and dest are arrays.
5118 generate_non_array_guard(load_object_klass(src), slow_region);
5119 generate_non_array_guard(load_object_klass(dest), slow_region);
5120
5121 // (2) src and dest arrays must have elements of the same BasicType
5122 // done at macro expansion or at Ideal transformation time
5123
5124 // (4) src_offset must not be negative.
5125 generate_negative_guard(src_offset, slow_region);
5126
5127 // (5) dest_offset must not be negative.
5128 generate_negative_guard(dest_offset, slow_region);
5129
5130 // (7) src_offset + length must not exceed length of src.
5131 generate_limit_guard(src_offset, length,
5132 load_array_length(src),
5133 slow_region);
5134
5135 // (8) dest_offset + length must not exceed length of dest.
5136 generate_limit_guard(dest_offset, length,
5137 load_array_length(dest),
5138 slow_region);
5139
5140 // (6) length must not be negative.
5141 // This is also checked in generate_arraycopy() during macro expansion, but
5142 // we also have to check it here for the case where the ArrayCopyNode will
5143 // be eliminated by Escape Analysis.
5144 if (EliminateAllocations) {
5145 generate_negative_guard(length, slow_region);
5146 negative_length_guard_generated = true;
5147 }
5148
5149 // (9) each element of an oop array must be assignable
5150 Node* src_klass = load_object_klass(src);
5151 Node* dest_klass = load_object_klass(dest);
5152 Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
5153
5154 if (not_subtype_ctrl != top()) {
5155 PreserveJVMState pjvms(this);
5156 set_control(not_subtype_ctrl);
5157 uncommon_trap(Deoptimization::Reason_intrinsic,
5158 Deoptimization::Action_make_not_entrant);
5159 assert(stopped(), "Should be stopped");
5160 }
5161 {
5162 PreserveJVMState pjvms(this);
5163 set_control(_gvn.transform(slow_region));
5164 uncommon_trap(Deoptimization::Reason_intrinsic,
5165 Deoptimization::Action_make_not_entrant);
5166 assert(stopped(), "Should be stopped");
5167 }
5168 }
5169
5170 arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
5171
5172 if (stopped()) {
5173 return true;
5174 }
5175
5176 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
5177 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5178 // so the compiler has a chance to eliminate them: during macro expansion,
5179 // we have to set their control (CastPP nodes are eliminated).
5180 load_object_klass(src), load_object_klass(dest),
5181 load_array_length(src), load_array_length(dest));
5182
5183 ac->set_arraycopy(validated);
5184
5185 Node* n = _gvn.transform(ac);
5186 if (n == ac) {
5187 ac->connect_outputs(this);
5188 } else {
5189 assert(validated, "shouldn't transform if all arguments not validated");
5190 set_all_memory(n);
5191 }
5192
5193 return true;
5194 }
5195
5196
5197 // Helper function which determines if an arraycopy immediately follows
5198 // an allocation, with no intervening tests or other escapes for the object.
5199 AllocateArrayNode*
5200 LibraryCallKit::tightly_coupled_allocation(Node* ptr,
5201 RegionNode* slow_region) {
5202 if (stopped()) return NULL; // no fast path
5203 if (C->AliasLevel() == 0) return NULL; // no MergeMems around
5204
5205 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5206 if (alloc == NULL) return NULL;
5207
5208 Node* rawmem = memory(Compile::AliasIdxRaw);
5209 // Is the allocation's memory state untouched?
5210 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5211 // Bail out if there have been raw-memory effects since the allocation.
5212 // (Example: There might have been a call or safepoint.)
5213 return NULL;
5214 }
5215 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5216 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5217 return NULL;
5218 }
5219
5220 // There must be no unexpected observers of this allocation.
5221 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5222 Node* obs = ptr->fast_out(i);
5223 if (obs != this->map()) {
5224 return NULL;
5225 }
5226 }
5227
5228 // This arraycopy must unconditionally follow the allocation of the ptr.
5229 Node* alloc_ctl = ptr->in(0);
5230 assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");
5231
5232 Node* ctl = control();
5233 while (ctl != alloc_ctl) {
5234 // There may be guards which feed into the slow_region.
5235 // Any other control flow means that we might not get a chance
5236 // to finish initializing the allocated object.
5237 if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
5238 IfNode* iff = ctl->in(0)->as_If();
5239 Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
5240 assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
5241 if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
5242 ctl = iff->in(0); // This test feeds the known slow_region.
5243 continue;
5244 }
5245 // One more try: Various low-level checks bottom out in
5246 // uncommon traps. If the debug-info of the trap omits
5247 // any reference to the allocation, as we've already
5248 // observed, then there can be no objection to the trap.
5249 bool found_trap = false;
5250 for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
5251 Node* obs = not_ctl->fast_out(j);
5252 if (obs->in(0) == not_ctl && obs->is_Call() &&
5253 (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5254 found_trap = true; break;
5255 }
5256 }
5257 if (found_trap) {
5258 ctl = iff->in(0); // This test feeds a harmless uncommon trap.
5259 continue;
5260 }
5261 }
5262 return NULL;
5263 }
5264
5265 // If we get this far, we have an allocation which immediately
5266 // precedes the arraycopy, and we can take over zeroing the new object.
5267 // The arraycopy will finish the initialization, and provide
5268 // a new control state to which we will anchor the destination pointer.
5269
5270 return alloc;
5271 }
5272
5273 //-------------inline_encodeISOArray-----------------------------------
5274 // encode char[] to byte[] in ISO_8859_1
5275 bool LibraryCallKit::inline_encodeISOArray() {
5276 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5277 // no receiver since it is static method
5278 Node *src = argument(0);
5279 Node *src_offset = argument(1);
5280 Node *dst = argument(2);
5281 Node *dst_offset = argument(3);
5282 Node *length = argument(4);
5283
5284 const Type* src_type = src->Value(&_gvn);
5285 const Type* dst_type = dst->Value(&_gvn);
5286 const TypeAryPtr* top_src = src_type->isa_aryptr();
5287 const TypeAryPtr* top_dest = dst_type->isa_aryptr();
5288 if (top_src == NULL || top_src->klass() == NULL ||
5289 top_dest == NULL || top_dest->klass() == NULL) {
5290 // failed array check
5291 return false;
5292 }
5293
5294 // Figure out the size and type of the elements we will be copying.
5295 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5296 BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5297 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5298 return false;
5299 }
5300
5301 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5302 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5303 // 'src_start' points to src array + scaled offset
5304 // 'dst_start' points to dst array + scaled offset
5305
5306 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5307 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
5308 enc = _gvn.transform(enc);
5309 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5310 set_memory(res_mem, mtype);
5311 set_result(enc);
5312 return true;
5313 }
5314
5315 //-------------inline_multiplyToLen-----------------------------------
5316 bool LibraryCallKit::inline_multiplyToLen() {
5317 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5318
5319 address stubAddr = StubRoutines::multiplyToLen();
5320 if (stubAddr == NULL) {
5321 return false; // Intrinsic's stub is not implemented on this platform
5322 }
5323 const char* stubName = "multiplyToLen";
5324
5325 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5326
5327 // no receiver because it is a static method
5328 Node* x = argument(0);
5329 Node* xlen = argument(1);
5330 Node* y = argument(2);
5331 Node* ylen = argument(3);
5332 Node* z = argument(4);
5333
5334 const Type* x_type = x->Value(&_gvn);
5335 const Type* y_type = y->Value(&_gvn);
5336 const TypeAryPtr* top_x = x_type->isa_aryptr();
5337 const TypeAryPtr* top_y = y_type->isa_aryptr();
5338 if (top_x == NULL || top_x->klass() == NULL ||
5339 top_y == NULL || top_y->klass() == NULL) {
5340 // failed array check
5341 return false;
5342 }
5343
5344 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5345 BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5346 if (x_elem != T_INT || y_elem != T_INT) {
5347 return false;
5348 }
5349
5350 // Set the original stack and the reexecute bit for the interpreter to reexecute
5351 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5352 // on the return from z array allocation in runtime.
5353 { PreserveReexecuteState preexecs(this);
5354 jvms()->set_should_reexecute(true);
5355
5356 Node* x_start = array_element_address(x, intcon(0), x_elem);
5357 Node* y_start = array_element_address(y, intcon(0), y_elem);
5358 // 'x_start' points to x array + scaled xlen
5359 // 'y_start' points to y array + scaled ylen
5360
5361 // Allocate the result array
5362 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5363 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5364 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5365
5366 IdealKit ideal(this);
5367
5368 #define __ ideal.
5369 Node* one = __ ConI(1);
5370 Node* zero = __ ConI(0);
5371 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5372 __ set(need_alloc, zero);
5373 __ set(z_alloc, z);
5374 __ if_then(z, BoolTest::eq, null()); {
5375 __ increment (need_alloc, one);
5376 } __ else_(); {
5377 // Update graphKit memory and control from IdealKit.
5378 sync_kit(ideal);
5379 Node* zlen_arg = load_array_length(z);
5380 // Update IdealKit memory and control from graphKit.
5381 __ sync_kit(this);
5382 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5383 __ increment (need_alloc, one);
5384 } __ end_if();
5385 } __ end_if();
5386
5387 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5388 // Update graphKit memory and control from IdealKit.
5389 sync_kit(ideal);
5390 Node * narr = new_array(klass_node, zlen, 1);
5391 // Update IdealKit memory and control from graphKit.
5392 __ sync_kit(this);
5393 __ set(z_alloc, narr);
5394 } __ end_if();
5395
5396 sync_kit(ideal);
5397 z = __ value(z_alloc);
5398 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5399 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5400 // Final sync IdealKit and GraphKit.
5401 final_sync(ideal);
5402 #undef __
5403
5404 Node* z_start = array_element_address(z, intcon(0), T_INT);
5405
5406 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5407 OptoRuntime::multiplyToLen_Type(),
5408 stubAddr, stubName, TypePtr::BOTTOM,
5409 x_start, xlen, y_start, ylen, z_start, zlen);
5410 } // original reexecute is set back here
5411
5412 C->set_has_split_ifs(true); // Has chance for split-if optimization
5413 set_result(z);
5414 return true;
5415 }
5416
5417 //-------------inline_squareToLen------------------------------------
5418 bool LibraryCallKit::inline_squareToLen() {
5419 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5420
5421 address stubAddr = StubRoutines::squareToLen();
5422 if (stubAddr == NULL) {
5423 return false; // Intrinsic's stub is not implemented on this platform
5424 }
5425 const char* stubName = "squareToLen";
5426
5427 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5428
5429 Node* x = argument(0);
5430 Node* len = argument(1);
5431 Node* z = argument(2);
5432 Node* zlen = argument(3);
5433
5434 const Type* x_type = x->Value(&_gvn);
5435 const Type* z_type = z->Value(&_gvn);
5436 const TypeAryPtr* top_x = x_type->isa_aryptr();
5437 const TypeAryPtr* top_z = z_type->isa_aryptr();
5438 if (top_x == NULL || top_x->klass() == NULL ||
5439 top_z == NULL || top_z->klass() == NULL) {
5440 // failed array check
5441 return false;
5442 }
5443
5444 BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5445 BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5446 if (x_elem != T_INT || z_elem != T_INT) {
5447 return false;
5448 }
5449
5450
5451 Node* x_start = array_element_address(x, intcon(0), x_elem);
5452 Node* z_start = array_element_address(z, intcon(0), z_elem);
5453
5454 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5455 OptoRuntime::squareToLen_Type(),
5456 stubAddr, stubName, TypePtr::BOTTOM,
5457 x_start, len, z_start, zlen);
5458
5459 set_result(z);
5460 return true;
5461 }
5462
5463 //-------------inline_mulAdd------------------------------------------
5464 bool LibraryCallKit::inline_mulAdd() {
5465 assert(UseMulAddIntrinsic, "not implemented on this platform");
5466
5467 address stubAddr = StubRoutines::mulAdd();
5468 if (stubAddr == NULL) {
5469 return false; // Intrinsic's stub is not implemented on this platform
5470 }
5471 const char* stubName = "mulAdd";
5472
5473 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5474
5475 Node* out = argument(0);
5476 Node* in = argument(1);
5477 Node* offset = argument(2);
5478 Node* len = argument(3);
5479 Node* k = argument(4);
5480
5481 const Type* out_type = out->Value(&_gvn);
5482 const Type* in_type = in->Value(&_gvn);
5483 const TypeAryPtr* top_out = out_type->isa_aryptr();
5484 const TypeAryPtr* top_in = in_type->isa_aryptr();
5485 if (top_out == NULL || top_out->klass() == NULL ||
5486 top_in == NULL || top_in->klass() == NULL) {
5487 // failed array check
5488 return false;
5489 }
5490
5491 BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5492 BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5493 if (out_elem != T_INT || in_elem != T_INT) {
5494 return false;
5495 }
5496
5497 Node* outlen = load_array_length(out);
5498 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5499 Node* out_start = array_element_address(out, intcon(0), out_elem);
5500 Node* in_start = array_element_address(in, intcon(0), in_elem);
5501
5502 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5503 OptoRuntime::mulAdd_Type(),
5504 stubAddr, stubName, TypePtr::BOTTOM,
5505 out_start,in_start, new_offset, len, k);
5506 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5507 set_result(result);
5508 return true;
5509 }
5510
5511 //-------------inline_montgomeryMultiply-----------------------------------
5512 bool LibraryCallKit::inline_montgomeryMultiply() {
5513 address stubAddr = StubRoutines::montgomeryMultiply();
5514 if (stubAddr == NULL) {
5515 return false; // Intrinsic's stub is not implemented on this platform
5516 }
5517
5518 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5519 const char* stubName = "montgomery_multiply";
5520
5521 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5522
5523 Node* a = argument(0);
5524 Node* b = argument(1);
5525 Node* n = argument(2);
5526 Node* len = argument(3);
5527 Node* inv = argument(4);
5528 Node* m = argument(6);
5529
5530 const Type* a_type = a->Value(&_gvn);
5531 const TypeAryPtr* top_a = a_type->isa_aryptr();
5532 const Type* b_type = b->Value(&_gvn);
5533 const TypeAryPtr* top_b = b_type->isa_aryptr();
5534 const Type* n_type = a->Value(&_gvn);
5535 const TypeAryPtr* top_n = n_type->isa_aryptr();
5536 const Type* m_type = a->Value(&_gvn);
5537 const TypeAryPtr* top_m = m_type->isa_aryptr();
5538 if (top_a == NULL || top_a->klass() == NULL ||
5539 top_b == NULL || top_b->klass() == NULL ||
5540 top_n == NULL || top_n->klass() == NULL ||
5541 top_m == NULL || top_m->klass() == NULL) {
5542 // failed array check
5543 return false;
5544 }
5545
5546 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5547 BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5548 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5549 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5550 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5551 return false;
5552 }
5553
5554 // Make the call
5555 {
5556 Node* a_start = array_element_address(a, intcon(0), a_elem);
5557 Node* b_start = array_element_address(b, intcon(0), b_elem);
5558 Node* n_start = array_element_address(n, intcon(0), n_elem);
5559 Node* m_start = array_element_address(m, intcon(0), m_elem);
5560
5561 Node* call = make_runtime_call(RC_LEAF,
5562 OptoRuntime::montgomeryMultiply_Type(),
5563 stubAddr, stubName, TypePtr::BOTTOM,
5564 a_start, b_start, n_start, len, inv, top(),
5565 m_start);
5566 set_result(m);
5567 }
5568
5569 return true;
5570 }
5571
5572 bool LibraryCallKit::inline_montgomerySquare() {
5573 address stubAddr = StubRoutines::montgomerySquare();
5574 if (stubAddr == NULL) {
5575 return false; // Intrinsic's stub is not implemented on this platform
5576 }
5577
5578 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5579 const char* stubName = "montgomery_square";
5580
5581 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5582
5583 Node* a = argument(0);
5584 Node* n = argument(1);
5585 Node* len = argument(2);
5586 Node* inv = argument(3);
5587 Node* m = argument(5);
5588
5589 const Type* a_type = a->Value(&_gvn);
5590 const TypeAryPtr* top_a = a_type->isa_aryptr();
5591 const Type* n_type = a->Value(&_gvn);
5592 const TypeAryPtr* top_n = n_type->isa_aryptr();
5593 const Type* m_type = a->Value(&_gvn);
5594 const TypeAryPtr* top_m = m_type->isa_aryptr();
5595 if (top_a == NULL || top_a->klass() == NULL ||
5596 top_n == NULL || top_n->klass() == NULL ||
5597 top_m == NULL || top_m->klass() == NULL) {
5598 // failed array check
5599 return false;
5600 }
5601
5602 BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5603 BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5604 BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5605 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5606 return false;
5607 }
5608
5609 // Make the call
5610 {
5611 Node* a_start = array_element_address(a, intcon(0), a_elem);
5612 Node* n_start = array_element_address(n, intcon(0), n_elem);
5613 Node* m_start = array_element_address(m, intcon(0), m_elem);
5614
5615 Node* call = make_runtime_call(RC_LEAF,
5616 OptoRuntime::montgomerySquare_Type(),
5617 stubAddr, stubName, TypePtr::BOTTOM,
5618 a_start, n_start, len, inv, top(),
5619 m_start);
5620 set_result(m);
5621 }
5622
5623 return true;
5624 }
5625
5626 //-------------inline_vectorizedMismatch------------------------------
5627 bool LibraryCallKit::inline_vectorizedMismatch() {
5628 assert(UseVectorizedMismatchIntrinsic, "not implementated on this platform");
5629
5630 address stubAddr = StubRoutines::vectorizedMismatch();
5631 if (stubAddr == NULL) {
5632 return false; // Intrinsic's stub is not implemented on this platform
5633 }
5634 const char* stubName = "vectorizedMismatch";
5635 int size_l = callee()->signature()->size();
5636 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5637
5638 Node* obja = argument(0);
5639 Node* aoffset = argument(1);
5640 Node* objb = argument(3);
5641 Node* boffset = argument(4);
5642 Node* length = argument(6);
5643 Node* scale = argument(7);
5644
5645 const Type* a_type = obja->Value(&_gvn);
5646 const Type* b_type = objb->Value(&_gvn);
5647 const TypeAryPtr* top_a = a_type->isa_aryptr();
5648 const TypeAryPtr* top_b = b_type->isa_aryptr();
5649 if (top_a == NULL || top_a->klass() == NULL ||
5650 top_b == NULL || top_b->klass() == NULL) {
5651 // failed array check
5652 return false;
5653 }
5654
5655 Node* call;
5656 jvms()->set_should_reexecute(true);
5657
5658 Node* obja_adr = make_unsafe_address(obja, aoffset);
5659 Node* objb_adr = make_unsafe_address(objb, boffset);
5660
5661 call = make_runtime_call(RC_LEAF,
5662 OptoRuntime::vectorizedMismatch_Type(),
5663 stubAddr, stubName, TypePtr::BOTTOM,
5664 obja_adr, objb_adr, length, scale);
5665
5666 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5667 set_result(result);
5668 return true;
5669 }
5670
5671 /**
5672 * Calculate CRC32 for byte.
5673 * int java.util.zip.CRC32.update(int crc, int b)
5674 */
5675 bool LibraryCallKit::inline_updateCRC32() {
5676 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5677 assert(callee()->signature()->size() == 2, "update has 2 parameters");
5678 // no receiver since it is static method
5679 Node* crc = argument(0); // type: int
5680 Node* b = argument(1); // type: int
5681
5682 /*
5683 * int c = ~ crc;
5684 * b = timesXtoThe32[(b ^ c) & 0xFF];
5685 * b = b ^ (c >>> 8);
5686 * crc = ~b;
5687 */
5688
5689 Node* M1 = intcon(-1);
5690 crc = _gvn.transform(new XorINode(crc, M1));
5691 Node* result = _gvn.transform(new XorINode(crc, b));
5692 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5693
5694 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5695 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5696 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5697 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5698
5699 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5700 result = _gvn.transform(new XorINode(crc, result));
5701 result = _gvn.transform(new XorINode(result, M1));
5702 set_result(result);
5703 return true;
5704 }
5705
5706 /**
5707 * Calculate CRC32 for byte[] array.
5708 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5709 */
5710 bool LibraryCallKit::inline_updateBytesCRC32() {
5711 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5712 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5713 // no receiver since it is static method
5714 Node* crc = argument(0); // type: int
5715 Node* src = argument(1); // type: oop
5716 Node* offset = argument(2); // type: int
5717 Node* length = argument(3); // type: int
5718
5719 const Type* src_type = src->Value(&_gvn);
5720 const TypeAryPtr* top_src = src_type->isa_aryptr();
5721 if (top_src == NULL || top_src->klass() == NULL) {
5722 // failed array check
5723 return false;
5724 }
5725
5726 // Figure out the size and type of the elements we will be copying.
5727 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5728 if (src_elem != T_BYTE) {
5729 return false;
5730 }
5731
5732 // 'src_start' points to src array + scaled offset
5733 Node* src_start = array_element_address(src, offset, src_elem);
5734
5735 // We assume that range check is done by caller.
5736 // TODO: generate range check (offset+length < src.length) in debug VM.
5737
5738 // Call the stub.
5739 address stubAddr = StubRoutines::updateBytesCRC32();
5740 const char *stubName = "updateBytesCRC32";
5741
5742 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5743 stubAddr, stubName, TypePtr::BOTTOM,
5744 crc, src_start, length);
5745 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5746 set_result(result);
5747 return true;
5748 }
5749
5750 /**
5751 * Calculate CRC32 for ByteBuffer.
5752 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5753 */
5754 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5755 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5756 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5757 // no receiver since it is static method
5758 Node* crc = argument(0); // type: int
5759 Node* src = argument(1); // type: long
5760 Node* offset = argument(3); // type: int
5761 Node* length = argument(4); // type: int
5762
5763 src = ConvL2X(src); // adjust Java long to machine word
5764 Node* base = _gvn.transform(new CastX2PNode(src));
5765 offset = ConvI2X(offset);
5766
5767 // 'src_start' points to src array + scaled offset
5768 Node* src_start = basic_plus_adr(top(), base, offset);
5769
5770 // Call the stub.
5771 address stubAddr = StubRoutines::updateBytesCRC32();
5772 const char *stubName = "updateBytesCRC32";
5773
5774 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5775 stubAddr, stubName, TypePtr::BOTTOM,
5776 crc, src_start, length);
5777 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5778 set_result(result);
5779 return true;
5780 }
5781
5782 //------------------------------get_table_from_crc32c_class-----------------------
5783 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5784 Node* table = load_field_from_object(NULL, "byteTable", "[I", /*is_exact*/ false, /*is_static*/ true, crc32c_class);
5785 assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5786
5787 return table;
5788 }
5789
5790 //------------------------------inline_updateBytesCRC32C-----------------------
5791 //
5792 // Calculate CRC32C for byte[] array.
5793 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5794 //
5795 bool LibraryCallKit::inline_updateBytesCRC32C() {
5796 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5797 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5798 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5799 // no receiver since it is a static method
5800 Node* crc = argument(0); // type: int
5801 Node* src = argument(1); // type: oop
5802 Node* offset = argument(2); // type: int
5803 Node* end = argument(3); // type: int
5804
5805 Node* length = _gvn.transform(new SubINode(end, offset));
5806
5807 const Type* src_type = src->Value(&_gvn);
5808 const TypeAryPtr* top_src = src_type->isa_aryptr();
5809 if (top_src == NULL || top_src->klass() == NULL) {
5810 // failed array check
5811 return false;
5812 }
5813
5814 // Figure out the size and type of the elements we will be copying.
5815 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5816 if (src_elem != T_BYTE) {
5817 return false;
5818 }
5819
5820 // 'src_start' points to src array + scaled offset
5821 Node* src_start = array_element_address(src, offset, src_elem);
5822
5823 // static final int[] byteTable in class CRC32C
5824 Node* table = get_table_from_crc32c_class(callee()->holder());
5825 Node* table_start = array_element_address(table, intcon(0), T_INT);
5826
5827 // We assume that range check is done by caller.
5828 // TODO: generate range check (offset+length < src.length) in debug VM.
5829
5830 // Call the stub.
5831 address stubAddr = StubRoutines::updateBytesCRC32C();
5832 const char *stubName = "updateBytesCRC32C";
5833
5834 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5835 stubAddr, stubName, TypePtr::BOTTOM,
5836 crc, src_start, length, table_start);
5837 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5838 set_result(result);
5839 return true;
5840 }
5841
5842 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5843 //
5844 // Calculate CRC32C for DirectByteBuffer.
5845 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5846 //
5847 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5848 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5849 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5850 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5851 // no receiver since it is a static method
5852 Node* crc = argument(0); // type: int
5853 Node* src = argument(1); // type: long
5854 Node* offset = argument(3); // type: int
5855 Node* end = argument(4); // type: int
5856
5857 Node* length = _gvn.transform(new SubINode(end, offset));
5858
5859 src = ConvL2X(src); // adjust Java long to machine word
5860 Node* base = _gvn.transform(new CastX2PNode(src));
5861 offset = ConvI2X(offset);
5862
5863 // 'src_start' points to src array + scaled offset
5864 Node* src_start = basic_plus_adr(top(), base, offset);
5865
5866 // static final int[] byteTable in class CRC32C
5867 Node* table = get_table_from_crc32c_class(callee()->holder());
5868 Node* table_start = array_element_address(table, intcon(0), T_INT);
5869
5870 // Call the stub.
5871 address stubAddr = StubRoutines::updateBytesCRC32C();
5872 const char *stubName = "updateBytesCRC32C";
5873
5874 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5875 stubAddr, stubName, TypePtr::BOTTOM,
5876 crc, src_start, length, table_start);
5877 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5878 set_result(result);
5879 return true;
5880 }
5881
5882 //------------------------------inline_updateBytesAdler32----------------------
5883 //
5884 // Calculate Adler32 checksum for byte[] array.
5885 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5886 //
5887 bool LibraryCallKit::inline_updateBytesAdler32() {
5888 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5889 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5890 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5891 // no receiver since it is static method
5892 Node* crc = argument(0); // type: int
5893 Node* src = argument(1); // type: oop
5894 Node* offset = argument(2); // type: int
5895 Node* length = argument(3); // type: int
5896
5897 const Type* src_type = src->Value(&_gvn);
5898 const TypeAryPtr* top_src = src_type->isa_aryptr();
5899 if (top_src == NULL || top_src->klass() == NULL) {
5900 // failed array check
5901 return false;
5902 }
5903
5904 // Figure out the size and type of the elements we will be copying.
5905 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5906 if (src_elem != T_BYTE) {
5907 return false;
5908 }
5909
5910 // 'src_start' points to src array + scaled offset
5911 Node* src_start = array_element_address(src, offset, src_elem);
5912
5913 // We assume that range check is done by caller.
5914 // TODO: generate range check (offset+length < src.length) in debug VM.
5915
5916 // Call the stub.
5917 address stubAddr = StubRoutines::updateBytesAdler32();
5918 const char *stubName = "updateBytesAdler32";
5919
5920 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5921 stubAddr, stubName, TypePtr::BOTTOM,
5922 crc, src_start, length);
5923 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5924 set_result(result);
5925 return true;
5926 }
5927
5928 //------------------------------inline_updateByteBufferAdler32---------------
5929 //
5930 // Calculate Adler32 checksum for DirectByteBuffer.
5931 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5932 //
5933 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5934 assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5935 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5936 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5937 // no receiver since it is static method
5938 Node* crc = argument(0); // type: int
5939 Node* src = argument(1); // type: long
5940 Node* offset = argument(3); // type: int
5941 Node* length = argument(4); // type: int
5942
5943 src = ConvL2X(src); // adjust Java long to machine word
5944 Node* base = _gvn.transform(new CastX2PNode(src));
5945 offset = ConvI2X(offset);
5946
5947 // 'src_start' points to src array + scaled offset
5948 Node* src_start = basic_plus_adr(top(), base, offset);
5949
5950 // Call the stub.
5951 address stubAddr = StubRoutines::updateBytesAdler32();
5952 const char *stubName = "updateBytesAdler32";
5953
5954 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5955 stubAddr, stubName, TypePtr::BOTTOM,
5956 crc, src_start, length);
5957
5958 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5959 set_result(result);
5960 return true;
5961 }
5962
5963 //----------------------------inline_reference_get----------------------------
5964 // public T java.lang.ref.Reference.get();
5965 bool LibraryCallKit::inline_reference_get() {
5966 const int referent_offset = java_lang_ref_Reference::referent_offset;
5967 guarantee(referent_offset > 0, "should have already been set");
5968
5969 // Get the argument:
5970 Node* reference_obj = null_check_receiver();
5971 if (stopped()) return true;
5972
5973 Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5974
5975 ciInstanceKlass* klass = env()->Object_klass();
5976 const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);
5977
5978 Node* no_ctrl = NULL;
5979 Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT, MemNode::unordered);
5980
5981 // Use the pre-barrier to record the value in the referent field
5982 pre_barrier(false /* do_load */,
5983 control(),
5984 NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5985 result /* pre_val */,
5986 T_OBJECT);
5987
5988 // Add memory barrier to prevent commoning reads from this field
5989 // across safepoint since GC can change its value.
5990 insert_mem_bar(Op_MemBarCPUOrder);
5991
5992 set_result(result);
5993 return true;
5994 }
5995
5996
5997 Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5998 bool is_exact=true, bool is_static=false,
5999 ciInstanceKlass * fromKls=NULL) {
6000 if (fromKls == NULL) {
6001 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6002 assert(tinst != NULL, "obj is null");
6003 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6004 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6005 fromKls = tinst->klass()->as_instance_klass();
6006 } else {
6007 assert(is_static, "only for static field access");
6008 }
6009 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6010 ciSymbol::make(fieldTypeString),
6011 is_static);
6012
6013 assert (field != NULL, "undefined field");
6014 if (field == NULL) return (Node *) NULL;
6015
6016 if (is_static) {
6017 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6018 fromObj = makecon(tip);
6019 }
6020
6021 // Next code copied from Parse::do_get_xxx():
6022
6023 // Compute address and memory type.
6024 int offset = field->offset_in_bytes();
6025 bool is_vol = field->is_volatile();
6026 ciType* field_klass = field->type();
6027 assert(field_klass->is_loaded(), "should be loaded");
6028 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6029 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6030 BasicType bt = field->layout_type();
6031
6032 // Build the resultant type of the load
6033 const Type *type;
6034 if (bt == T_OBJECT) {
6035 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6036 } else {
6037 type = Type::get_const_basic_type(bt);
6038 }
6039
6040 if (support_IRIW_for_not_multiple_copy_atomic_cpu && is_vol) {
6041 insert_mem_bar(Op_MemBarVolatile); // StoreLoad barrier
6042 }
6043 // Build the load.
6044 MemNode::MemOrd mo = is_vol ? MemNode::acquire : MemNode::unordered;
6045 Node* loadedField = make_load(NULL, adr, type, bt, adr_type, mo, LoadNode::DependsOnlyOnTest, is_vol);
6046 // If reference is volatile, prevent following memory ops from
6047 // floating up past the volatile read. Also prevents commoning
6048 // another volatile read.
6049 if (is_vol) {
6050 // Memory barrier includes bogus read of value to force load BEFORE membar
6051 insert_mem_bar(Op_MemBarAcquire, loadedField);
6052 }
6053 return loadedField;
6054 }
6055
6056 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6057 bool is_exact = true, bool is_static = false,
6058 ciInstanceKlass * fromKls = NULL) {
6059 if (fromKls == NULL) {
6060 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6061 assert(tinst != NULL, "obj is null");
6062 assert(tinst->klass()->is_loaded(), "obj is not loaded");
6063 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6064 fromKls = tinst->klass()->as_instance_klass();
6065 }
6066 else {
6067 assert(is_static, "only for static field access");
6068 }
6069 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6070 ciSymbol::make(fieldTypeString),
6071 is_static);
6072
6073 assert(field != NULL, "undefined field");
6074 assert(!field->is_volatile(), "not defined for volatile fields");
6075
6076 if (is_static) {
6077 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6078 fromObj = makecon(tip);
6079 }
6080
6081 // Next code copied from Parse::do_get_xxx():
6082
6083 // Compute address and memory type.
6084 int offset = field->offset_in_bytes();
6085 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6086
6087 return adr;
6088 }
6089
6090 //------------------------------inline_aescrypt_Block-----------------------
6091 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6092 address stubAddr = NULL;
6093 const char *stubName;
6094 assert(UseAES, "need AES instruction support");
6095
6096 switch(id) {
6097 case vmIntrinsics::_aescrypt_encryptBlock:
6098 stubAddr = StubRoutines::aescrypt_encryptBlock();
6099 stubName = "aescrypt_encryptBlock";
6100 break;
6101 case vmIntrinsics::_aescrypt_decryptBlock:
6102 stubAddr = StubRoutines::aescrypt_decryptBlock();
6103 stubName = "aescrypt_decryptBlock";
6104 break;
6105 }
6106 if (stubAddr == NULL) return false;
6107
6108 Node* aescrypt_object = argument(0);
6109 Node* src = argument(1);
6110 Node* src_offset = argument(2);
6111 Node* dest = argument(3);
6112 Node* dest_offset = argument(4);
6113
6114 // (1) src and dest are arrays.
6115 const Type* src_type = src->Value(&_gvn);
6116 const Type* dest_type = dest->Value(&_gvn);
6117 const TypeAryPtr* top_src = src_type->isa_aryptr();
6118 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6119 assert (top_src != NULL && top_src->klass() != NULL && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6120
6121 // for the quick and dirty code we will skip all the checks.
6122 // we are just trying to get the call to be generated.
6123 Node* src_start = src;
6124 Node* dest_start = dest;
6125 if (src_offset != NULL || dest_offset != NULL) {
6126 assert(src_offset != NULL && dest_offset != NULL, "");
6127 src_start = array_element_address(src, src_offset, T_BYTE);
6128 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6129 }
6130
6131 // now need to get the start of its expanded key array
6132 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6133 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6134 if (k_start == NULL) return false;
6135
6136 if (Matcher::pass_original_key_for_aes()) {
6137 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6138 // compatibility issues between Java key expansion and SPARC crypto instructions
6139 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6140 if (original_k_start == NULL) return false;
6141
6142 // Call the stub.
6143 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6144 stubAddr, stubName, TypePtr::BOTTOM,
6145 src_start, dest_start, k_start, original_k_start);
6146 } else {
6147 // Call the stub.
6148 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6149 stubAddr, stubName, TypePtr::BOTTOM,
6150 src_start, dest_start, k_start);
6151 }
6152
6153 return true;
6154 }
6155
6156 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6157 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6158 address stubAddr = NULL;
6159 const char *stubName = NULL;
6160
6161 assert(UseAES, "need AES instruction support");
6162
6163 switch(id) {
6164 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6165 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6166 stubName = "cipherBlockChaining_encryptAESCrypt";
6167 break;
6168 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6169 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6170 stubName = "cipherBlockChaining_decryptAESCrypt";
6171 break;
6172 }
6173 if (stubAddr == NULL) return false;
6174
6175 Node* cipherBlockChaining_object = argument(0);
6176 Node* src = argument(1);
6177 Node* src_offset = argument(2);
6178 Node* len = argument(3);
6179 Node* dest = argument(4);
6180 Node* dest_offset = argument(5);
6181
6182 // (1) src and dest are arrays.
6183 const Type* src_type = src->Value(&_gvn);
6184 const Type* dest_type = dest->Value(&_gvn);
6185 const TypeAryPtr* top_src = src_type->isa_aryptr();
6186 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6187 assert (top_src != NULL && top_src->klass() != NULL
6188 && top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6189
6190 // checks are the responsibility of the caller
6191 Node* src_start = src;
6192 Node* dest_start = dest;
6193 if (src_offset != NULL || dest_offset != NULL) {
6194 assert(src_offset != NULL && dest_offset != NULL, "");
6195 src_start = array_element_address(src, src_offset, T_BYTE);
6196 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6197 }
6198
6199 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6200 // (because of the predicated logic executed earlier).
6201 // so we cast it here safely.
6202 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6203
6204 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6205 if (embeddedCipherObj == NULL) return false;
6206
6207 // cast it to what we know it will be at runtime
6208 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6209 assert(tinst != NULL, "CBC obj is null");
6210 assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6211 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6212 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6213
6214 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6215 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6216 const TypeOopPtr* xtype = aklass->as_instance_type();
6217 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6218 aescrypt_object = _gvn.transform(aescrypt_object);
6219
6220 // we need to get the start of the aescrypt_object's expanded key array
6221 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6222 if (k_start == NULL) return false;
6223
6224 // similarly, get the start address of the r vector
6225 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
6226 if (objRvec == NULL) return false;
6227 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6228
6229 Node* cbcCrypt;
6230 if (Matcher::pass_original_key_for_aes()) {
6231 // on SPARC we need to pass the original key since key expansion needs to happen in intrinsics due to
6232 // compatibility issues between Java key expansion and SPARC crypto instructions
6233 Node* original_k_start = get_original_key_start_from_aescrypt_object(aescrypt_object);
6234 if (original_k_start == NULL) return false;
6235
6236 // Call the stub, passing src_start, dest_start, k_start, r_start, src_len and original_k_start
6237 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6238 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6239 stubAddr, stubName, TypePtr::BOTTOM,
6240 src_start, dest_start, k_start, r_start, len, original_k_start);
6241 } else {
6242 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6243 cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6244 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6245 stubAddr, stubName, TypePtr::BOTTOM,
6246 src_start, dest_start, k_start, r_start, len);
6247 }
6248
6249 // return cipher length (int)
6250 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6251 set_result(retvalue);
6252 return true;
6253 }
6254
6255 //------------------------------inline_counterMode_AESCrypt-----------------------
6256 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6257 assert(UseAES, "need AES instruction support");
6258 if (!UseAESCTRIntrinsics) return false;
6259
6260 address stubAddr = NULL;
6261 const char *stubName = NULL;
6262 if (id == vmIntrinsics::_counterMode_AESCrypt) {
6263 stubAddr = StubRoutines::counterMode_AESCrypt();
6264 stubName = "counterMode_AESCrypt";
6265 }
6266 if (stubAddr == NULL) return false;
6267
6268 Node* counterMode_object = argument(0);
6269 Node* src = argument(1);
6270 Node* src_offset = argument(2);
6271 Node* len = argument(3);
6272 Node* dest = argument(4);
6273 Node* dest_offset = argument(5);
6274
6275 // (1) src and dest are arrays.
6276 const Type* src_type = src->Value(&_gvn);
6277 const Type* dest_type = dest->Value(&_gvn);
6278 const TypeAryPtr* top_src = src_type->isa_aryptr();
6279 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6280 assert(top_src != NULL && top_src->klass() != NULL &&
6281 top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6282
6283 // checks are the responsibility of the caller
6284 Node* src_start = src;
6285 Node* dest_start = dest;
6286 if (src_offset != NULL || dest_offset != NULL) {
6287 assert(src_offset != NULL && dest_offset != NULL, "");
6288 src_start = array_element_address(src, src_offset, T_BYTE);
6289 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6290 }
6291
6292 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6293 // (because of the predicated logic executed earlier).
6294 // so we cast it here safely.
6295 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6296 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6297 if (embeddedCipherObj == NULL) return false;
6298 // cast it to what we know it will be at runtime
6299 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6300 assert(tinst != NULL, "CTR obj is null");
6301 assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6302 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6303 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6304 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6305 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6306 const TypeOopPtr* xtype = aklass->as_instance_type();
6307 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6308 aescrypt_object = _gvn.transform(aescrypt_object);
6309 // we need to get the start of the aescrypt_object's expanded key array
6310 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6311 if (k_start == NULL) return false;
6312 // similarly, get the start address of the r vector
6313 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B", /*is_exact*/ false);
6314 if (obj_counter == NULL) return false;
6315 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6316
6317 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B", /*is_exact*/ false);
6318 if (saved_encCounter == NULL) return false;
6319 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6320 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6321
6322 Node* ctrCrypt;
6323 if (Matcher::pass_original_key_for_aes()) {
6324 // no SPARC version for AES/CTR intrinsics now.
6325 return false;
6326 }
6327 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6328 ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6329 OptoRuntime::counterMode_aescrypt_Type(),
6330 stubAddr, stubName, TypePtr::BOTTOM,
6331 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6332
6333 // return cipher length (int)
6334 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6335 set_result(retvalue);
6336 return true;
6337 }
6338
6339 //------------------------------get_key_start_from_aescrypt_object-----------------------
6340 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6341 #if defined(PPC64) || defined(S390)
6342 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6343 // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6344 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6345 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6346 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I", /*is_exact*/ false);
6347 assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6348 if (objSessionK == NULL) {
6349 return (Node *) NULL;
6350 }
6351 Node* objAESCryptKey = load_array_element(control(), objSessionK, intcon(0), TypeAryPtr::OOPS);
6352 #else
6353 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
6354 #endif // PPC64
6355 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6356 if (objAESCryptKey == NULL) return (Node *) NULL;
6357
6358 // now have the array, need to get the start address of the K array
6359 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6360 return k_start;
6361 }
6362
6363 //------------------------------get_original_key_start_from_aescrypt_object-----------------------
6364 Node * LibraryCallKit::get_original_key_start_from_aescrypt_object(Node *aescrypt_object) {
6365 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "lastKey", "[B", /*is_exact*/ false);
6366 assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6367 if (objAESCryptKey == NULL) return (Node *) NULL;
6368
6369 // now have the array, need to get the start address of the lastKey array
6370 Node* original_k_start = array_element_address(objAESCryptKey, intcon(0), T_BYTE);
6371 return original_k_start;
6372 }
6373
6374 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6375 // Return node representing slow path of predicate check.
6376 // the pseudo code we want to emulate with this predicate is:
6377 // for encryption:
6378 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6379 // for decryption:
6380 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6381 // note cipher==plain is more conservative than the original java code but that's OK
6382 //
6383 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6384 // The receiver was checked for NULL already.
6385 Node* objCBC = argument(0);
6386
6387 // Load embeddedCipher field of CipherBlockChaining object.
6388 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6389
6390 // get AESCrypt klass for instanceOf check
6391 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6392 // will have same classloader as CipherBlockChaining object
6393 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6394 assert(tinst != NULL, "CBCobj is null");
6395 assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6396
6397 // we want to do an instanceof comparison against the AESCrypt class
6398 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6399 if (!klass_AESCrypt->is_loaded()) {
6400 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6401 Node* ctrl = control();
6402 set_control(top()); // no regular fast path
6403 return ctrl;
6404 }
6405 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6406
6407 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6408 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6409 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6410
6411 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6412
6413 // for encryption, we are done
6414 if (!decrypting)
6415 return instof_false; // even if it is NULL
6416
6417 // for decryption, we need to add a further check to avoid
6418 // taking the intrinsic path when cipher and plain are the same
6419 // see the original java code for why.
6420 RegionNode* region = new RegionNode(3);
6421 region->init_req(1, instof_false);
6422 Node* src = argument(1);
6423 Node* dest = argument(4);
6424 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6425 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6426 Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6427 region->init_req(2, src_dest_conjoint);
6428
6429 record_for_igvn(region);
6430 return _gvn.transform(region);
6431 }
6432
6433 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6434 // Return node representing slow path of predicate check.
6435 // the pseudo code we want to emulate with this predicate is:
6436 // for encryption:
6437 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6438 // for decryption:
6439 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6440 // note cipher==plain is more conservative than the original java code but that's OK
6441 //
6442
6443 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6444 // The receiver was checked for NULL already.
6445 Node* objCTR = argument(0);
6446
6447 // Load embeddedCipher field of CipherBlockChaining object.
6448 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
6449
6450 // get AESCrypt klass for instanceOf check
6451 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6452 // will have same classloader as CipherBlockChaining object
6453 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6454 assert(tinst != NULL, "CTRobj is null");
6455 assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6456
6457 // we want to do an instanceof comparison against the AESCrypt class
6458 ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6459 if (!klass_AESCrypt->is_loaded()) {
6460 // if AESCrypt is not even loaded, we never take the intrinsic fast path
6461 Node* ctrl = control();
6462 set_control(top()); // no regular fast path
6463 return ctrl;
6464 }
6465
6466 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6467 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6468 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6469 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6470 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6471
6472 return instof_false; // even if it is NULL
6473 }
6474
6475 //------------------------------inline_ghash_processBlocks
6476 bool LibraryCallKit::inline_ghash_processBlocks() {
6477 address stubAddr;
6478 const char *stubName;
6479 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6480
6481 stubAddr = StubRoutines::ghash_processBlocks();
6482 stubName = "ghash_processBlocks";
6483
6484 Node* data = argument(0);
6485 Node* offset = argument(1);
6486 Node* len = argument(2);
6487 Node* state = argument(3);
6488 Node* subkeyH = argument(4);
6489
6490 Node* state_start = array_element_address(state, intcon(0), T_LONG);
6491 assert(state_start, "state is NULL");
6492 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
6493 assert(subkeyH_start, "subkeyH is NULL");
6494 Node* data_start = array_element_address(data, offset, T_BYTE);
6495 assert(data_start, "data is NULL");
6496
6497 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6498 OptoRuntime::ghash_processBlocks_Type(),
6499 stubAddr, stubName, TypePtr::BOTTOM,
6500 state_start, subkeyH_start, data_start, len);
6501 return true;
6502 }
6503
6504 //------------------------------inline_sha_implCompress-----------------------
6505 //
6506 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6507 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6508 //
6509 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6510 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6511 //
6512 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6513 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6514 //
6515 bool LibraryCallKit::inline_sha_implCompress(vmIntrinsics::ID id) {
6516 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6517
6518 Node* sha_obj = argument(0);
6519 Node* src = argument(1); // type oop
6520 Node* ofs = argument(2); // type int
6521
6522 const Type* src_type = src->Value(&_gvn);
6523 const TypeAryPtr* top_src = src_type->isa_aryptr();
6524 if (top_src == NULL || top_src->klass() == NULL) {
6525 // failed array check
6526 return false;
6527 }
6528 // Figure out the size and type of the elements we will be copying.
6529 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6530 if (src_elem != T_BYTE) {
6531 return false;
6532 }
6533 // 'src_start' points to src array + offset
6534 Node* src_start = array_element_address(src, ofs, src_elem);
6535 Node* state = NULL;
6536 address stubAddr;
6537 const char *stubName;
6538
6539 switch(id) {
6540 case vmIntrinsics::_sha_implCompress:
6541 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6542 state = get_state_from_sha_object(sha_obj);
6543 stubAddr = StubRoutines::sha1_implCompress();
6544 stubName = "sha1_implCompress";
6545 break;
6546 case vmIntrinsics::_sha2_implCompress:
6547 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6548 state = get_state_from_sha_object(sha_obj);
6549 stubAddr = StubRoutines::sha256_implCompress();
6550 stubName = "sha256_implCompress";
6551 break;
6552 case vmIntrinsics::_sha5_implCompress:
6553 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6554 state = get_state_from_sha5_object(sha_obj);
6555 stubAddr = StubRoutines::sha512_implCompress();
6556 stubName = "sha512_implCompress";
6557 break;
6558 default:
6559 fatal_unexpected_iid(id);
6560 return false;
6561 }
6562 if (state == NULL) return false;
6563
6564 // Call the stub.
6565 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::sha_implCompress_Type(),
6566 stubAddr, stubName, TypePtr::BOTTOM,
6567 src_start, state);
6568
6569 return true;
6570 }
6571
6572 //------------------------------inline_digestBase_implCompressMB-----------------------
6573 //
6574 // Calculate SHA/SHA2/SHA5 for multi-block byte[] array.
6575 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6576 //
6577 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6578 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6579 "need SHA1/SHA256/SHA512 instruction support");
6580 assert((uint)predicate < 3, "sanity");
6581 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6582
6583 Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6584 Node* src = argument(1); // byte[] array
6585 Node* ofs = argument(2); // type int
6586 Node* limit = argument(3); // type int
6587
6588 const Type* src_type = src->Value(&_gvn);
6589 const TypeAryPtr* top_src = src_type->isa_aryptr();
6590 if (top_src == NULL || top_src->klass() == NULL) {
6591 // failed array check
6592 return false;
6593 }
6594 // Figure out the size and type of the elements we will be copying.
6595 BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6596 if (src_elem != T_BYTE) {
6597 return false;
6598 }
6599 // 'src_start' points to src array + offset
6600 Node* src_start = array_element_address(src, ofs, src_elem);
6601
6602 const char* klass_SHA_name = NULL;
6603 const char* stub_name = NULL;
6604 address stub_addr = NULL;
6605 bool long_state = false;
6606
6607 switch (predicate) {
6608 case 0:
6609 if (UseSHA1Intrinsics) {
6610 klass_SHA_name = "sun/security/provider/SHA";
6611 stub_name = "sha1_implCompressMB";
6612 stub_addr = StubRoutines::sha1_implCompressMB();
6613 }
6614 break;
6615 case 1:
6616 if (UseSHA256Intrinsics) {
6617 klass_SHA_name = "sun/security/provider/SHA2";
6618 stub_name = "sha256_implCompressMB";
6619 stub_addr = StubRoutines::sha256_implCompressMB();
6620 }
6621 break;
6622 case 2:
6623 if (UseSHA512Intrinsics) {
6624 klass_SHA_name = "sun/security/provider/SHA5";
6625 stub_name = "sha512_implCompressMB";
6626 stub_addr = StubRoutines::sha512_implCompressMB();
6627 long_state = true;
6628 }
6629 break;
6630 default:
6631 fatal("unknown SHA intrinsic predicate: %d", predicate);
6632 }
6633 if (klass_SHA_name != NULL) {
6634 // get DigestBase klass to lookup for SHA klass
6635 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6636 assert(tinst != NULL, "digestBase_obj is not instance???");
6637 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6638
6639 ciKlass* klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6640 assert(klass_SHA->is_loaded(), "predicate checks that this class is loaded");
6641 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6642 return inline_sha_implCompressMB(digestBase_obj, instklass_SHA, long_state, stub_addr, stub_name, src_start, ofs, limit);
6643 }
6644 return false;
6645 }
6646 //------------------------------inline_sha_implCompressMB-----------------------
6647 bool LibraryCallKit::inline_sha_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_SHA,
6648 bool long_state, address stubAddr, const char *stubName,
6649 Node* src_start, Node* ofs, Node* limit) {
6650 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_SHA);
6651 const TypeOopPtr* xtype = aklass->as_instance_type();
6652 Node* sha_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6653 sha_obj = _gvn.transform(sha_obj);
6654
6655 Node* state;
6656 if (long_state) {
6657 state = get_state_from_sha5_object(sha_obj);
6658 } else {
6659 state = get_state_from_sha_object(sha_obj);
6660 }
6661 if (state == NULL) return false;
6662
6663 // Call the stub.
6664 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6665 OptoRuntime::digestBase_implCompressMB_Type(),
6666 stubAddr, stubName, TypePtr::BOTTOM,
6667 src_start, state, ofs, limit);
6668 // return ofs (int)
6669 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6670 set_result(result);
6671
6672 return true;
6673 }
6674
6675 //------------------------------get_state_from_sha_object-----------------------
6676 Node * LibraryCallKit::get_state_from_sha_object(Node *sha_object) {
6677 Node* sha_state = load_field_from_object(sha_object, "state", "[I", /*is_exact*/ false);
6678 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA/SHA2");
6679 if (sha_state == NULL) return (Node *) NULL;
6680
6681 // now have the array, need to get the start address of the state array
6682 Node* state = array_element_address(sha_state, intcon(0), T_INT);
6683 return state;
6684 }
6685
6686 //------------------------------get_state_from_sha5_object-----------------------
6687 Node * LibraryCallKit::get_state_from_sha5_object(Node *sha_object) {
6688 Node* sha_state = load_field_from_object(sha_object, "state", "[J", /*is_exact*/ false);
6689 assert (sha_state != NULL, "wrong version of sun.security.provider.SHA5");
6690 if (sha_state == NULL) return (Node *) NULL;
6691
6692 // now have the array, need to get the start address of the state array
6693 Node* state = array_element_address(sha_state, intcon(0), T_LONG);
6694 return state;
6695 }
6696
6697 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6698 // Return node representing slow path of predicate check.
6699 // the pseudo code we want to emulate with this predicate is:
6700 // if (digestBaseObj instanceof SHA/SHA2/SHA5) do_intrinsic, else do_javapath
6701 //
6702 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6703 assert(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics,
6704 "need SHA1/SHA256/SHA512 instruction support");
6705 assert((uint)predicate < 3, "sanity");
6706
6707 // The receiver was checked for NULL already.
6708 Node* digestBaseObj = argument(0);
6709
6710 // get DigestBase klass for instanceOf check
6711 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6712 assert(tinst != NULL, "digestBaseObj is null");
6713 assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6714
6715 const char* klass_SHA_name = NULL;
6716 switch (predicate) {
6717 case 0:
6718 if (UseSHA1Intrinsics) {
6719 // we want to do an instanceof comparison against the SHA class
6720 klass_SHA_name = "sun/security/provider/SHA";
6721 }
6722 break;
6723 case 1:
6724 if (UseSHA256Intrinsics) {
6725 // we want to do an instanceof comparison against the SHA2 class
6726 klass_SHA_name = "sun/security/provider/SHA2";
6727 }
6728 break;
6729 case 2:
6730 if (UseSHA512Intrinsics) {
6731 // we want to do an instanceof comparison against the SHA5 class
6732 klass_SHA_name = "sun/security/provider/SHA5";
6733 }
6734 break;
6735 default:
6736 fatal("unknown SHA intrinsic predicate: %d", predicate);
6737 }
6738
6739 ciKlass* klass_SHA = NULL;
6740 if (klass_SHA_name != NULL) {
6741 klass_SHA = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_SHA_name));
6742 }
6743 if ((klass_SHA == NULL) || !klass_SHA->is_loaded()) {
6744 // if none of SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6745 Node* ctrl = control();
6746 set_control(top()); // no intrinsic path
6747 return ctrl;
6748 }
6749 ciInstanceKlass* instklass_SHA = klass_SHA->as_instance_klass();
6750
6751 Node* instofSHA = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass_SHA)));
6752 Node* cmp_instof = _gvn.transform(new CmpINode(instofSHA, intcon(1)));
6753 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6754 Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6755
6756 return instof_false; // even if it is NULL
6757 }
6758
6759 //-------------inline_fma-----------------------------------
6760 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6761 Node *a = NULL;
6762 Node *b = NULL;
6763 Node *c = NULL;
6764 Node* result = NULL;
6765 switch (id) {
6766 case vmIntrinsics::_fmaD:
6767 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6768 // no receiver since it is static method
6769 a = round_double_node(argument(0));
6770 b = round_double_node(argument(2));
6771 c = round_double_node(argument(4));
6772 result = _gvn.transform(new FmaDNode(control(), a, b, c));
6773 break;
6774 case vmIntrinsics::_fmaF:
6775 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6776 a = argument(0);
6777 b = argument(1);
6778 c = argument(2);
6779 result = _gvn.transform(new FmaFNode(control(), a, b, c));
6780 break;
6781 default:
6782 fatal_unexpected_iid(id); break;
6783 }
6784 set_result(result);
6785 return true;
6786 }
6787
6788 bool LibraryCallKit::inline_profileBoolean() {
6789 Node* counts = argument(1);
6790 const TypeAryPtr* ary = NULL;
6791 ciArray* aobj = NULL;
6792 if (counts->is_Con()
6793 && (ary = counts->bottom_type()->isa_aryptr()) != NULL
6794 && (aobj = ary->const_oop()->as_array()) != NULL
6795 && (aobj->length() == 2)) {
6796 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
6797 jint false_cnt = aobj->element_value(0).as_int();
6798 jint true_cnt = aobj->element_value(1).as_int();
6799
6800 if (C->log() != NULL) {
6801 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
6802 false_cnt, true_cnt);
6803 }
6804
6805 if (false_cnt + true_cnt == 0) {
6806 // According to profile, never executed.
6807 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6808 Deoptimization::Action_reinterpret);
6809 return true;
6810 }
6811
6812 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
6813 // is a number of each value occurrences.
6814 Node* result = argument(0);
6815 if (false_cnt == 0 || true_cnt == 0) {
6816 // According to profile, one value has been never seen.
6817 int expected_val = (false_cnt == 0) ? 1 : 0;
6818
6819 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
6820 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6821
6822 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
6823 Node* fast_path = _gvn.transform(new IfTrueNode(check));
6824 Node* slow_path = _gvn.transform(new IfFalseNode(check));
6825
6826 { // Slow path: uncommon trap for never seen value and then reexecute
6827 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
6828 // the value has been seen at least once.
6829 PreserveJVMState pjvms(this);
6830 PreserveReexecuteState preexecs(this);
6831 jvms()->set_should_reexecute(true);
6832
6833 set_control(slow_path);
6834 set_i_o(i_o());
6835
6836 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
6837 Deoptimization::Action_reinterpret);
6838 }
6839 // The guard for never seen value enables sharpening of the result and
6840 // returning a constant. It allows to eliminate branches on the same value
6841 // later on.
6842 set_control(fast_path);
6843 result = intcon(expected_val);
6844 }
6845 // Stop profiling.
6846 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
6847 // By replacing method body with profile data (represented as ProfileBooleanNode
6848 // on IR level) we effectively disable profiling.
6849 // It enables full speed execution once optimized code is generated.
6850 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
6851 C->record_for_igvn(profile);
6852 set_result(profile);
6853 return true;
6854 } else {
6855 // Continue profiling.
6856 // Profile data isn't available at the moment. So, execute method's bytecode version.
6857 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
6858 // is compiled and counters aren't available since corresponding MethodHandle
6859 // isn't a compile-time constant.
6860 return false;
6861 }
6862 }
6863
6864 bool LibraryCallKit::inline_isCompileConstant() {
6865 Node* n = argument(0);
6866 set_result(n->is_Con() ? intcon(1) : intcon(0));
6867 return true;
6868 }